Integrating Robotic Bilateral Axillo-Breast Approach Thyroidectomy with Molecular Diagnostics and Artificial Intelligence in Thyroid Cancer Care

Abstract

Thyroid cancer is one of the most prevalent cancers in the world, accounting for the increased sensitivity of diagnostic assessments, the environment, and extensive imaging. While open thyroidectomy still forms the base surgical treatment of differentiated thyroid carcinoma (DTC), its visible scarring and postoperative morbidity have mandatorily propelled a longitudinal shift to minimally invasive techniques. Robotic thyroidectomy using the bilateral axillo-breast approach (BABA) is an emerging transformative technique that combines robotic precision and visualization with oncologic efficacy and superior cosmetic outcomes to enhance recovery. Molecular diagnostics such as next-generation sequencing (NGS) and microRNA classifiers have transformed preoperative planning. These tools aid in accurate risk stratification, enabling clinicians to determine the course of surgery and avoid overtreatment. Artificial intelligence (AI) enhances precision medicine by improving nodule classification, predicting surgical risks, providing intraoperative navigation guidance, and supporting postoperative histopathological evaluations. Despite these innovations, cost, accessibility, and ethical governance issues highlight persistent challenges. This review consolidates the current state of robotic-assisted surgery for thyroid cancer with molecular profiling and AI. It proposes future aims and strides toward precision surgery that is accessible worldwide.

INTRODUCTION

Thyroid cancer is one of the most frequently diagnosed cancers worldwide. Recent developments in diagnostics, as well as environmental and genetic factors, are potential contributors (Jin et al., 2025). According to GLOBOCAN estimations, thyroid cancer diagnosis cases in 2020 reached 586,000 with 43,646 which are 3.0 percent and 0.4 percent of total cancer cases and deaths, respectively (Sung et al., 2021). Although it ranks as the ninth most common cancer globally, the incidence exhibits a striking gender disparity, with women affected approximately three times more frequently than men (Lim et al., 2017).

The global distribution of thyroid cancer reveals notable heterogeneity. In South Korea, age-standardized incidence rates (ASRs) have surpassed 60 per 100,000 women and 15 per 100,000 men, primarily attributed to the introduction of widespread national screening programs in the early 2000s (Majeed et al., 2024). This surge often termed the “thyroid cancer epidemic,” led to a near-tripling of incidence rates over a decade, with a peak observed in 2012 prior to a modest decline following public health interventions aimed at mitigating over-diagnosis (Grimm, 2022). Similarly, in the United States, thyroid cancer incidence has risen nearly 300% since the 1970s, driven predominantly by increased detection of small, localized papillary thyroid carcinomas (PTCs), frequently identified incidentally through imaging modalities employed for unrelated clinical concerns (Reed and Mallick, 2017).

Furthermore, countries in Europe, such as Italy, France, and Spain, and high-resource countries like Australia and New Zealand, have documented parallel increases due to enhanced access to technology that provides diagnoses (Gao et al., 2025). Unlike most of the continent, South Asia, and Latin America report substantially lower incidence rates. However, these figures are likely underestimated due to limited access to advanced diagnostic technologies and weak health system infrastructures, which obscure the actual burden of disease (Alagoz et al., 2024). In low- and middle-income countries (LMICs), deficiencies in timely diagnosis, accessible healthcare services, and effective treatment contribute to more frequent advanced-stage presentations and higher mortality rates from thyroid cancer (Kitahara and Sosa, 2020; Li et al., 2020).

Amid shifting epidemiological patterns, surgical resection remains the cornerstone of treatment for differentiated thyroid cancer (DTC), the most common and typically indolent subtype. For over a century, Kocher’s transverse incision on the anterior neck was the standard approach for thyroidectomy (Christoforides et al., 2018). While effective for direct access to the thyroid gland, this technique was associated with several postoperative complications, including hypesthesia, paresthesia, and dysphagia. Additionally, the visible scar often raised concerns about cosmetic outcomes, particularly among younger patients and women, negatively impacting psychological well-being and self-esteem (Sajisevi et al., 2022). This concern has spurred the development of remote-access techniques to eliminate visible scarring while maintaining surgical efficacy (Tufano et al., 2022).

In addition, early endoscopic methods presented significant challenges for surgeons, including limited visibility, inflexible tools, and poor ergonomics. These difficulties led to introducing robotics, which solved many of the issues posed by previous techniques (Anuwong, 2024). In the last two decades, there has been a radical shift in the approach to thyroid surgery from open classic surgeries with a visible cervical incision to less invasive, scar-free, aesthetically optimized procedures while still being oncologically sound (Lee et al., 2022). The pioneering work of Gagner and Hüscher in the 1990s—introducing endoscopic head and neck surgery and endoscopic thyroidectomy, respectively—catalyzed this paradigm shift (Procopio et al., 2023). Subsequent advancements in high-definition endoscopy and robotic surgical systems facilitated the development of remote-access thyroidectomy via routes such as the axillary, breast, anterior chest, postauricular facelift, and transoral approaches. This evolution has ushered in a new era of scarless endocrine surgery (Sun et al., 2022).

Among these innovations, robotic thyroidectomy BABA (Bilateral Axillo-breast approach) is the most advanced (Kang et al., 2024). BABA integrates both technical precision and technical excellence. Using this surgical technique, surgeons make four small symmetry incisions—axillary and two areolar—to access the gland, achieving optimal concealment and preserving the neck’s natural contour while gaining wide access to the surgical field (Al Juhani et al., 2025). This approach preserves the neck’s natural contour while providing wide access to the surgical field. With robotic systems such as the da Vinci Surgical System, enhanced three-dimensional views of the surgical field, improved instrument articulation, and helm-controlled manipulators allow for safer dissection around critical structures like the recurrent laryngeal nerve and parathyroid glands, greatly diminishing postoperative complications (Park and Kim, 2024). This technique delivers a balanced outcome—minimizing visible scarring and complications while maximizing operative control and patient satisfaction. (Paek et al., 2022).

Although surgical advances like BABA laparoscopy consider the functional and aesthetic aspects of thyroid cancer treatment, contemporaneous developments in molecular diagnostics have revolutionized the approach to evaluating thyroid nodules preoperatively (Shin and Bae, 2022). Next-generation sequencing (NGS), mutational profiling, and gene expression classifiers enhance risk stratification, improving individualized preoperative planning (Krubaa et al., 2024). By classifying thyroid cancer into indolent and aggressive categories, molecular diagnostics enable the customizing of the extent of thyroidectomy to the biology, thereby decreasing overtreatment and improving long-term results. Incorporating molecular profiling into clinical pathways is an important aspect of precision endocrine surgery alongside advancements in minimally invasive approaches, profiling, and molecular medicine (Liu et al., 2023; Parpounas and Constantinides, 2023).

Nonetheless, these technological innovations also create problem areas in cost, accessibility, and clinical uniformity. The same applies to robotic thyroidectomy and molecular diagnostics; both require considerable expenditure and healthcare resources, stunting broad adoption, particularly in LMICs (Sipos and Ringel, 2023). Furthermore, differences in diagnostic systems and surgical techniques emphasize the importance of multicenter research, real-world data on clinical outcomes, and uniform clinical guidelines for evidence-based management (Feyzi and Sheykhloo, 2023).

Integrating surgery and endocrinology with interventional radiology, pathology, molecular biology, and survivorship care constitutes a comprehensive, multidisciplinary roadmap for the future of thyroid cancer surgery (Kim, 2025). In this evolving paradigm, the incorporation of Artificial Intelligence (AI) plays a pivotal role in enhancing diagnostic accuracy, surgical planning, risk stratification, and personalized treatment algorithms (Giorgini et al., 2024). Moreover, innovative surgical techniques such as the Bilateral Axillo-Breast Approach (BABA) offer cosmetically favorable, minimally invasive alternatives that align with patient-centric values, particularly regarding aesthetic outcomes. A unified, patient-focused approach—encompassing precision anatomy, advanced imaging, equitable access to cutting-edge technology, and standardized success metrics—must be developed to ensure that healthcare systems worldwide benefit from these advances (Toro-Tobon et al., 2023). Establishing such a model will not only improve oncologic outcomes but also enhance patients’ quality of life, satisfaction with postoperative cervical appearance, and long-term structural survivorship in individuals with thyroid malignancies (Taha et al., 2024).

In this review, we examine the evolution of thyroidectomy from open to robotic and scarless approaches, the clinical advantages and limitations of BABA robotic thyroidectomy, and the transformative role of molecular diagnostics in operative planning. Our aim is to provide a comprehensive perspective on the shift toward precision, scarless, and patient-centered management in endocrine surgery.

TYPES AND CLINICAL FEATURES OF THYROID CANCER

Recent advances in research and clinical practice have refined the classification and characterization of thyroid cancers, enabling more personalized care.

Differentiated thyroid cancer

Differentiated thyroid cancers are the most frequently encountered type of thyroid cancer, constituting nearly 90-95% of all thyroid malignancies. Such cancers develop from the follicular cells of the thyroid gland, which synthesize the thyroid hormones (Chen et al., 2023). Moreover, thyroid cancers, differentiated in particular, tend to have good prognoses and are generally manageable, particularly when diagnosed in the early stages. The primary subtypes of differentiated thyroid cancer are Papillary Thyroid Carcinoma (PTC) and Follicular Thyroid Carcinoma (FTC) (Means et al., 2019).

Papillary Thyroid Carcinoma (PTC): Papillary Thyroid Carcinoma (PTC) is the most common type of thyroid malignancy, representing about 89.1% of thyroid cancers diagnosed, and has a rising incidence rate globally, especially in females. Primary contributors to such malignancy include ionizing radiation, imbalances of dietary iodine, obesity, hormonal changes, and environmental pollutants (Schmidbauer et al., 2017). Microscopically, the defining feature of PTC is the presence of papillae. Each papilla is made up of layers of neoplastic cells around a fibrovascular core (Pozdeyev et al., 2018). Typical cellular histomorphology includes large and clear nuclei with finely granular chromatin, sometimes referred to as ground-glass or “Orphan Annie-eyed” nuclei, which exhibit nuclear grooving and have inclusions within the nucleus. Calcified clusters of cells, termed psammoma bodies, which are thought to result from necrosis of papillary structures, are also commonly observed (Fig. 1) (Prete et al., 2020).

Fig. 1.

Histopathological and Molecular Features of Thyroid Cancer Subtypes. (a) Key molecular alterations in each subtype: PTC – BRAF V600E, RET/PTC rearrangements; FTC – RAS mutations, PAX8-PPARG fusions; ATC – TP53, TERT promoter mutations; MTC – RET point mutations. (b) Diagrammatic representation of Papillary (PTC) (Lin et al., 2011), Follicular (FTC) (Bhattacharya et al., 2023), Anaplastic (ATC) (Ragazzi et al., 2014), and Medullary (MTC) thyroid cancers (Thomas et al., 2019) showing key morphological features: PTC – papillary structures with characteristic nuclear changes; FTC – follicular architecture; ATC – highly undifferentiated cells; MTC – spindle-shaped cells with amyloid deposits.

From a clinical perspective, PTC usually presents as a thyroid nodule without pain and may be associated with lymph node enlargement in the neck in some individuals. In patients with PTC, the prognosis is usually good, but active surveillance for recurrences as well as the development of secondary primary tumors is critical. Genetically, BRAF V600E mutations are the most common alterations associated with PTC and, along with BRAF, are the mutations most positively correlated with the clinical and pathological features of PTC (Asa et al., 2025).

Surgical management includes total thyroidectomy for higher-risk or larger tumors and lobectomy for lower-risk lesions. Selective lymph node dissection is performed for lymphadenopathy or lymph node metastases based on clinical or radiological evaluation. Post-operative radioactive iodine (RAI) ablation and thyroid hormone suppression are given for intermediate and high-risk patients (Ulisse et al., 2021). Surveillance with neck ultrasonography, serum thyroglobulin, and imaging facilitates the earliest possible detection of recurrence and risk adaptive management tailored to the individual while preserving the generally good prognosis with papillary thyroid carcinoma (PTC) (Kuenstner et al., 2025).

Follicular Thyroid Carcinoma (FTC): FTC is the third most prevalent subtype of thyroid cancer and accounts for approximately 10 percent of all thyroid tumors (Baloch and LiVolsi, 2018). It is more common in regions lacking sufficient iodine and is typically manifested as a solitary and painless thyroid nodule. While FTC does not usually metastasize to the lymph nodes, it tends to metastasize to bones, the lungs, and the liver via hematogenous routes (Fig. 1) (Shen et al., 2024). FTC is primarily diagnosed using FNA biopsy and is subsequently confirmed by histopathology post-surgery. Compared to PTC, FTC is associated with a poorer prognosis; however, the prognosis is significantly improved with timely treatment (Gronlund et al., 2021).

The histological features of FTC are quite heterogeneous, ranging from well-differentiated follicular patterns to poorly differentiated patterns characterized by severe nuclear atypia, lack of follicular structures, extensive capsular or vascular invasion, and a solid growth pattern. These latter changes are associated with a poor prognosis. A follicular carcinoma is differentiated from a benign follicular adenoma solely based on invasion of the capsule and/or vascular structures (Fagin et al., 2023). Further, FTC is classified as minimally invasive, encapsulated, angioinvasive, and widely invasive based on the degree of invasion. Relatively common mutations in the RAS gene with FTC are associated with its distinctive biological features and are linked to the tumor biology of FTC at a molecular level (Ahmadi and Landa, 2024).

Thyroid FTCs require surgical management as the first line of treatment. For aggressive tumors over 4 cm, total thyroidectomy is indicated, while lobectomy suffices for small, low-risk lesions (Boucai et al., 2024). Intermediate- and high-risk patients require postoperative radioactive iodine ablation and TSH-suppressive therapy (Park et al., 2022). These patients are managed with serum thyroglobulin levels and imaging. In the case of FTCs that are resistant to RAI therapy or have metastasized, there are effective targeted options using tyrosine kinase inhibitors that remain as viable therapeutic options (Machens et al., 2022).

Poorly differentiated and anaplastic thyroid cancer

Poorly Differentiated Thyroid Cancer (PDTC) represents an intermediate entity between well-differentiated and anaplastic forms of malignancy, and derives from thyroid hormone-producing tissues. Furthermore, PDTC exhibits some features with well-differentiated thyroid malignancies, such as papillary and follicular carcinoma, and the aggressive anaplastic thyroid cancer (ATC) (Fig. 1) (Seok et al., 2025). Its histology shows solid, trabecular, and insular growth patterns that partially lose differentiation characteristics, which can aid in disguise diagnosis. Piermattei et al. (2025) suggest that “MAPK and PI3K-AKT signaling pathways” are responsible for the overactive proliferation of PDTC key oncogenes; therefore, mutations in crucial oncogenes, such as BRAF, RAS, and TERT promoter regions, are essential to tumor survival (Tong et al., 2022). The frequency of molecular mutations in PDTC is not uniform across studies, which may reflect not just differences in the patient population but also differences in the sensitivity of the methodology of genetic characterization, such as next-generation versus Sanger sequencing. These methodological differences markedly influence the detection of such mutations (Dagher et al., 2025).

Anaplastic Thyroid Cancer (ATC) is the most advanced form of thyroid cancer and is known for its rapid progression and poor prognosis. Clinically, ATC is marked by thyroid gland hyperplasia, severe compressive symptoms like dyspnea and dysphagia, and early metastasis to the lungs, bones, and regional lymph nodes (Cleere et al., 2024). At the genomic level, ATC is characterized by a highly unstable genome with numerous mutations that contribute to its aggressive nature. TP53 mutations, which occur in 80% of cases, permit unregulated tumor proliferation by disrupting cell cycle and apoptosis mechanisms (Abdalla et al., 2024). TERT promoter mutations, which occur in 40% to 70% of ATC, are associated with the immortalization of cells due to telomerase activation, particularly in the presence of BRAF V600E mutations that activate the MAPK pathway. Contributing RAS mutations, including NRAS, KRAS, and HRAS, activate MAPK and PI3K-AKT pathways, increasing cellular proliferation and survival of the cell (Fig. 1) (Subbiah et al., 2022). Other changes, such as PIK3CA mutations, PTEN loss, and deletions of CDKN2A/B, foster resistance to apoptosis and dysregulation of the cell cycle. Clinically, some of the most intriguing—but rare—fusions of the genes ALK, NTRK, and RET could enable personalized treatment strategies (Suda and Mitsudomi, 2020).

The molecular profile of ATC differentiates it from PDTC, which is characterized by partially preserved differentiation and some responsiveness to treatment. At the same time, ATC is almost completely dedifferentiated and unresponsive to treatment (Zeng et al., 2024). In the management of anaplastic thyroid cancer, patients undergo BRAF V600E mutation testing and staging. Surgical resection is performed for the controllable disease, followed by treatment with specific BRAF kinase inhibitors for those harboring BRAF V600E mutations (da Silva et al., 2023). The remaining patients received targeted radiation therapy and adjuvant cytotoxic chemotherapy. Patients often present with distant metastases, an aggressive course of the disease, and local invasion to the trachea or blood vessels, which renders surgical resection impossible. The cancer mortality rate is close to 100% and some patients may be deemed too risky for any intervention beyond conservative surgical techniques aimed at palliative relief (Rao et al., 2023).

Medullary thyroid cancer (MTC)

Medullary thyroid carcinoma (MTC) is an uncommon and undocumented type of malignancy in thyroid tissues for it develops from the C-cells or parafollicular cells of the thyroid gland which synthesizes the hormone calcitonin (Censi et al., 2023). Clinically, MTC is biochemically distinct from differentiated thyroid cancers (DTC) and is thus considered a separated disease for it constitutes approximately 3 – 5 % of thyroid cancers (West et al., 2025). MTC clinically may manifest as a solitary thyroid nodule, which is a well-defined and mobile structure, or as multicentric disease in hereditary forms. Cervical lymphadenopathy is common as a diagnostic accompaniment due to the tumor’s propensity for spreading from the subcutaneous tissue to the regional lymphatic chains (Fig. 1) (Grani et al., 2018). Some patients may have hormonal secretion and present with systemic calcitonin and a peptide called carcinoembryonic antigen (CEA), which is associated with the malignancy and results in flushing and diarrhea (Accardo et al., 2017).

The pathogenesis of MTC is closely associated with germline alterations in the RET proto-oncogene found on chromosome 10q11.2. These changes result in tyrosine kinase signaling C-cell proliferation, leading to increased cell division and tumor formation (Randle et al., 2017). In sporadic MTC, somatic mutations in RET, particularly RET M918T, are common and associated with more aggressive disease, increased metastatic spread, and an overall worse prognosis (Shaghaghi et al., 2022). MTC demonstrates a range of histological architectures with trabecular, solid, and glandular MTC. As Machens et al. (2022) describe, the diagnosis remains iodinated in the tumor with calcitonin, CEA, chromogranin A, and occasionally synaptophysin, confirming the diagnosis via immunohistochemistry. For molecular screening in hereditary cases, the identification of RET mutations is essential to inform risk evaluation and for tailoring prevention strategies, including active surveillance and prophylactic thyroidectomy.

Characteristics of pediatric and adolescent thyroid cancer

Thyroid cancers are infrequent within the pediatric and adolescent populations, yet they have unique epidemiological, clinical, and molecular features in comparison to thyroid cancers in adults. Pediatric thyroid cancers are mostly diagnosed as differentiated thyroid cancers, primarily as papillary thyroid carcinoma, followed by follicular thyroid carcinoma (FTC) (Stosic et al., 2021). In comparison to adult cases, pediatric thyroid cancers tend to have a more aggressive clinical course with higher rates of lymph node metastasis, extrathyroidal extension, and distant metastasis, notably to the lungs, although the prognosis remains good in the long term (Pan et al., 2017).

Within the context of thyroid cancers, the morphology and genetics of pediatric and adult cancers diverge considerably. Unlike adult PTC, pediatric thyroid cancers have a lower prevalence of BRAF V600E mutations, while exhibiting higher frequencies of RET/PTC rearrangements. These rearrangements are notably increased in children with a history of radiation exposure. The resultant RET/PTC fusions lead to the constitutive activation of the MAPK pathway, which enhances cell proliferation and survival (Fig. 1) (Sapuppo et al., 2021). In pediatric cases, the alteration of additional genetics including NTRK fusions, RAS mutations, and PAX8/PPARγ rearrangements are recognized, contributing to the tumorigenesis and clinical behaviour (Kim et al., 2022). Of particular interest, pediatric thyroid cancers are rarely associated with TERT promoter mutations, which in adults are strongly linked to aggressive disease. This phenomenon is hypothesized to contribute to the favorable prognosis in children, despite presenting with extensive disease (Tian et al., 2023).

In clinical practice, children and adolescents present with a neck mass, cervical lymphadenopathy, or other imaging lymphadenopathy for a completely different clinical concern. It is peculiar how aggressive the presentation is, juxtaposed against the high responsiveness to treatment, which is due to the preserved differentiation and radiosensitivity of the tumors (Remiker et al., 2019). Pediatric PTCs are also characterized by the presence of advanced multifocality and more aggressive patterns of spread in the thyroid and regional lymphatics, leading to the need for more aggressive operation such as total thyroidectomy with prophylactic central neck dissection in some cases (Araque et al., 2017).

ADVANCES IN ROBOTIC THYROIDECTOMY: THE BILATERAL AXILLO-BREAST APPROACH (BABA)

Evolution from open surgery to remote access techniques

Open thyroidectomy (OT) has been practiced for over a century as the definitive modality of treatment for thyroid cancer. Surgical access was reliably gained through Kocher’s transverse cervical incision, though it did incur a surgical cosmetic scar on the anterior neck (Gupta and Kataria, 2024). Considering that the majority of patients who have thyroid cancer are young women and the prognosis is usually good, thyroid cancer’s cosmetically unfriendly postoperative aesthetic outcomes have gained prominence alongside clinical considerations. Traditional surgical approaches often result in noticeable scars that can negatively impact patients’ psychosocial dynamics, self-esteem, and overall satisfaction, leading to marked dissatisfaction (Xu et al., 2024).

There was a paradigm change in thyroid surgery with the swift evolution of minimally invasive surgical techniques in the mid-2000s (Qin et al., 2023). Developments in endoscopic tools and imaging technology made it possible to perform more sophisticated procedures through more minor and less conspicuous openings. Surgeons initially applied these techniques to abdominal and thoracic surgery, extending their use to cervical surgeries. Endocrine surgeons pioneered various methods of remote-access thyroidectomy aimed at reducing cosmetic neck scars, such as the transaxillary approach (TAA), transoral vestibular approach (TOETVA), and bilateral axillo-breast approach (BABA) (Zhang et al., 2023a).

Developments such as BABA have marked critical milestones, and developments that followed have enhanced results. The BABA (bilateral axillo-breast approach) is a technique that Chae and coworkers developed at Seoul National University Hospital (SNUH) (Chae et al., 2022). BABA altered ABBA by refining incision neck placement and incision port access to make pelvic cuts greater than 2.5 cm above the pubis. BABA is effective and is building on its origins as an endoscopic procedure (Liang et al., 2021).

While endoscopic techniques are helpful, the lack of an unobstructed view and the low angle of approach severely limit instrument maneuverability. The BABA robotic thyroidectomy (BABA RoT) is the outcome of merging BABA with robotic techniques (Ludwig et al., 2023). The adoption of robotics brought better instrument maneuverability with 3D visualization coupled with tremor suppression and more precise, smoother movements. In-depth investigations have carried out more research endorsing BABA RoT as a safe and effective technique for scarless thyroidectomy with better cosmetic and functional results since the first clinical report (Albazee et al., 2023).

Surgical anatomy and access design of robotic BABA thyroidectomy

Robotic bilateral axillo-breast approach (BABA) thyroidectomy emphasizes balanced and secure access to the thyroid gland sans any scarring on the neck (Kang et al., 2022). This approach uses the principles of robotic surgery together with remote access techniques to spatially and cosmetically optimize the outcomes of the thyroidectomy (Ouyang et al., 2022). The surgical team operates a well-defined stepwise approach to the procedure, which is systematic and can be repeated as detailed below:

Preparation, positioning, and draping: Before an operation, an appropriate strategy and patient selection need to be made with care. The procedure is suitable for patients with small to moderately sized thyroid nodules or those with low—to intermediate-risk differentiated thyroid cancer. Under general anesthesia, patients are positioned supine with slight neck extension. Both arms are abducted to approximately 90 degrees to expose the axillary region. The chest and neck are widely prepped and draped to maintain sterility and permit unimpeded surgical access (Choi et al., 2024).

Drawing guidelines and injection: Anatomic landmarks are marked preoperatively on the chest and neck to guide incision placement and flap elevation. An epinephrine-mixed saline solution is injected subcutaneously along the planned dissection plane. This technique reduces intraoperative bleeding, facilitates hydro dissection, and creates a well-defined working space (Kim et al., 2018).

Skin incision, blunt dissection, and port insertion:

Four small incisions are made in discrete locations:

a) Two axillary incisions (approximately 0.8 cm each), hidden in the natural skin folds.

b) Two circumareolar incisions (0.8 cm on the left and 1.2 cm on the right).

Blunt dissection begins with the creation of subcutaneous tunnels toward the neck. Flap formation is then completed using sharp dissection with an energy device. Ports are inserted through these incisions, establishing channels for robotic instruments and visualization (Warfield et al., 2020).

Robot docking and complete flap elevation: After placing the four ports, the surgical team adjusts the operating bed to a reverse Trendelenburg position at approximately 20-30 degrees. This positioning allows gravitational retraction of the soft tissues , thereby enhancing exposure of the surgical field during robotic dissection (Meibner et al., 2021). Next, the surgical team prepares the robotic system. The central columns of the robotic carts and the camera arm are carefully aligned in a straight trajectory with the designated areolar camera port. The team then sequentially docks and connects the robotic arms to each access port (Wang et al., 2025b). They activate the surgical console and introduce the robotic instruments:

Camera port: A three-dimensional high-definition endoscopic camera is inserted through the correct areolar port to magnify the visualization of the operative field.

Left areolar port: A monopolar electrocautery or ultrasonic shear is inserted for dissection and vessel sealing.

Bilateral axillary ports: Graspers, typically ProGrasp and Maryland forceps, are inserted to facilitate tissue manipulation and exposure (Kim et al., 2015).

With the robotic system fully docked and the instruments engaged, dissection of the subcutaneous space proceeds. Using the robotic instruments, the surgeon completes the elevation of the flap in a controlled and bloodless manner. The dissection extends superiorly to the level of the thyroid cartilage, inferiorly to approximately 2 cm below the clavicle, and laterally beyond the medial border of the sternocleidomastoid (SCM) muscle. This broad and symmetrical exposure allows optimal access to the thyroid lobes and central compartment (Yu et al., 2024).

Recent surgical refinements have favored the creation of flaps in the subfascial plane rather than the conventional subplatysmal layer. Emerging evidence suggests that subfascial dissection reduces postoperative adhesions, potentially improving long-term patient comfort and cosmetic outcomes. When using this approach, ligation of the anterior jugular veins becomes necessary (Dabas et al., 2024). This procedure is safely performed near the sternal notch, utilizing ultrasonic shears or bipolar coagulation through Maryland forceps, which minimizes the risk of bleeding and ensures a clear operative field.Through these carefully coordinated steps, robot docking and complete flap elevation are achieved safely and effectively, providing the surgeon with optimal visualization and access to proceed with thyroidectomy (Woods et al., 2024).

The Key Steps of Dissection

Flap elevation and strap muscle retraction: Following the establishment of the subcutaneous working space and CO₂ insufflation, dissection proceeds in the subplatysmal plane toward the thyroid cartilage. The surgical team takes care to maintain flap integrity to avoid postoperative complications such as skin dimpling or seroma. Upon exposure of the strap muscles, the surgeon performs a vertical midline division. The robotic arms retract these muscles laterally, providing broad and symmetrical exposure of the anterior thyroid gland, which is essential for safe dissection, particularly during bilateral procedures (Cheng et al., 2024).

Identification and preservation of critical structures: Dissection begins at the inferior pole of the thyroid lobe. The inferior thyroid veins are carefully ligated using robotic bipolar or ultrasonic instruments to minimize bleeding and maintain a clear operative field. The recurrent laryngeal nerve (RLN) is identified early, typically at its entry point near the tracheoesophageal groove. The robotic platform provides magnified, three-dimensional visualization, which aids in precise dissection around the nerve to prevent traction or thermal injury. Preserving the parathyroid glands and their vascular supply is prioritized simultaneously. When necessary, devascularized parathyroid glands are auto-transplanted to prevent postoperative hypocalcemia (Zhang et al., 2024b).

Thyroid lobe mobilization and resection: Once the inferior pole and adjacent critical structures are secured, the thyroid lobe is mobilized superiorly. If present, the middle thyroid veins are divided, and superior pole dissection is performed with particular attention to the external branch of the superior laryngeal nerve (EBSLN), which is at risk during ligation of the superior thyroid vessels. Careful handling in this area prevents voice-related complications. The thyroid is detached from the Berry ligament, a key anatomical landmark closely associated with the RLN, requiring delicate and controlled dissection. Following complete mobilization, a lobectomy or total thyroidectomy is performed, depending on the surgical plan. The procedure concludes with unilateral lobe removal in lobectomy; in total thyroidectomy, the surgeon transects the isthmus and continues dissection on the contralateral lobe (Kurganov et al., 2024).

Isthmus division and contralateral lobectomy: For total thyroidectomy, the isthmus is divided at its attachment to the trachea using energy devices, completing ipsilateral resection. The contralateral lobe is approached using identical steps, maintaining the same principles of nerve preservation and vascular control. A thorough inspection ensures hemostasis and preservation of critical structures before specimen retrieval (Tewari and Kaur, 2025).

Specimen retrieval and closure: After thyroid removal, the specimen is placed in an endoscopic retrieval bag and extracted through the slightly enlarged right areolar incision. The surgeon verifies hemostasis and may place a suction drain if necessary. CO₂ insufflation is released, and subcutaneous tissues are allowed to reapproximate naturally. All incisions are closed using absorbable sutures to ensure optimal healing and minimal scarring. This approach avoids visible cervical scarring on the anterior neck, greatly enhancing cosmetic outcomes (Agcaoglu et al., 2024).

Postoperative management: The robotic approach minimizes the surgical field’s impact on the tissues, so patients report low postoperative pain and discomfort levels. Most patients can ingest food or fluids approximately 2 h after extubation. Clinicians typically discharge patients approximately 24-48 h after surgery, highlighting the procedure’s minimally invasive nature and rapid postoperative recovery (Lee and Yi, 2025).

Standard postoperative protocols include hypocalcemia screening, which entails close monitoring of serum calcium levels, especially with considerable dissection or total thyroidectomy. In addition, the medical team monitors laryngeal function through a simple voice check or laryngoscopy to preserve the recurrent laryngeal nerve. These actions aid in confirming the absence of immediate complications and determining the appropriate postoperative care plan (Zhou et al., 2024).

INTEGRATION OF MOLECULAR INSIGHTS INTO ROBOTIC SURGICAL DECISION-MAKING

Recent advances in cancer genomics have completely redefined the understanding of thyroid cancer. Researchers previously considered thyroid carcinoma to exhibit slow progression and limited biological variability; however, current evidence reveals it as a heterogeneous malignancy characterized by distinct genetic alterations. These changes have profound biological and clinical impacts (Tjota et al., 2024). Moreover, significant triaging and cultivation of thyroid tumors has demonstrated that they contain particular drivers of oncogenesis that give rise to and maintain cancer. These drivers influence the shape of the tumor, its ability to spread, and its response to treatment (Jose et al., 2024).

Recent advancements in robotic surgery for thyroidectomies, specifically BABA, have shifted the landscape of thyroid cancer surgery due to the incorporation of molecular oncology in its management. In surgical decision making, in the past, heavily relied on radiologic and histopathologic evaluation, genomic data of the tumor and its biology, as well as the cancer’s behavior and treatment responses, have provided critical insights vital to the diagnosis and treatment planning. Acuna-Ruiz et al. (2023) emphasizes how the incorporation of integrative oncology approaches alongside robotic surgical advancements has deepened the understanding of thyroid cancer at the molecular and biological levels, thereby improving clinical practice.

Tumor-specific genomic alterations have led to the creation of genomic biomarkers aimed at enhancing the diagnosis, assessment of prognosis, and formulation of targeted therapy. Unlike traditional histopathology approaches, and in line with the evolving principles of precision oncology, molecular profiling ensures accurate tumor subtyping, thereby informing and optimizing the surgical and adjuvant treatment plan. As highlighted by Agarwal et al. (2021), there is a paradigm shift with an emphasis on effective surgery in conjunction with precision medicine guided by molecular biology, with the ultimate goal of enhancing the clinical outcomes of thyroid cancer patients.

The malignant transformation of thyroid glands is strongly associated with the presence of specific oncogenes, which lead to cell growth and survival without control. The primary oncogenes implicated in thyroid cancer have been BRAF, RAS and RET, each with its specific role in the disease’s mechanism of action (Yoo et al., 2016). BRAF is a serine/threonine kinase and a member of the RAF family. BRAF is functionally considered a keystone of the MAPK/ERK pathway, cellular metabolism, growth, and apoptosis. Out of all thyroid malignancies, papillary thyroid carcinoma (PTC) is the most frequent one to have BRAF mutations, particularly the V600E variant, which is a substitute of valine with glutamic acid at codon 600. This alteration causes chronic activation of the kinase, which in turn enhances the signaling of the MAPK pathway, resulting in tumorigenesis, dedifferentiation, and loss of responsiveness to iodine-131 (radioactive iodine) therapy (Nannini et al., 2023).

The V600E mutation in BRAF is clinically linked with aggressive disease characteristics such as extrathyroidal extension, metastasis to lymph nodes, disease recurrence, and decreased overall survival. In other subtypes of thyroid cancer, BRAF mutations are infrequent; however, BRAF mutations in papillary thyroid carcinoma (PTC) are of great biological and therapeutic importance. Besides BRAF mutations fueling the MAPK pathway, they may also activate other signaling pathways like PI3K/AKT and WNT, which promote tumor progression (Fig. 2) (Pozdeyev et al., 2018).

Fig. 2.

Key molecular pathways in thyroid cancer are driven by genetic alterations and tumor-promoting cytokines. Dysregulation of the MAPK and PI3K/AKT pathways is primarily caused by mutations in BRAF, RAS, RET/PTC, NTRK, PTEN, PIK3CA, and AKT1. Additional contributors include PAX8-PPARG fusions, Wnt/β-catenin activation, TP53 inactivation, TERT promoter mutations, and pro-inflammatory cytokines that enhance tumor progression (Liu et al., 2021).

RAS genes, including NRAS, HRAS, and KRAS, encode small GTPases that control major pathways such as MAPK/ERK and PI3K/AKT, which regulate cell growth, death, and differentiation. Approximately 20-30% of thyroid cancers, most commonly found in follicular thyroid carcinoma (FTC) and the follicular variant of PTC, possess RAS mutations, which activate their pathways constitutively (Chen et al., 2018). RAS mutation-driven tumors are often associated with slower clinical progression than observed in BRAF-mutant PTC. RAS mutations, however, may in some cases—especially alongside other genetic changes—accelerate progression toward poorly differentiated or anaplastic thyroid carcinoma, illustrating their potential to act as oncogenes (Duan et al., 2019). It has been difficult to directly target RAS mutations because of their associated weak activation, commensurate multilogical connections, and the fact that they are shut down within numerous signaling pathways. Even so, RAS status is still crucial in the context of diagnostic optimization, stratified analysis, and in the case of new therapies designed to target components that act downstream to signaling pathways (Fig. 2) (Yakushina et al., 2018).

RET functions as a receptor tyrosine kinase whose chromosomal rearrangement–induced activation via RET/PTC fusions leads to oncogenesis via constitutive MAPK/ERK pathway activation. Such fusions are observed in approximately 10-20% of PTCs, especially in post-radiation exposure and pediatrics. RET/PTC rearrangements are associated with maintained differentiated cellular features, which, alongside tumor growth, are linked with a favorable prognosis and enhanced radioiodine sensitivity. On the other hand, RET-driven tumors may demonstrate greater aggressiveness in some histological variants or advanced disease stages (Liu et al., 2021).

Clinical decision-making for preoperative surgical planning of a procedure is becoming more precise with the integration of molecular profiling. Patients identified with high-risk genetic alterations may need a prophylactic central neck dissection, even when radiologic imaging shows no evidence of nodal metastasis. This is to minimize the chance of locoregional recurrence of the disease and to optimize postoperative surveillance. On the other hand, patients identified with low-risk profiles may require more conservative surgical treatment, with exemption from lymphadenectomy, thus reducing the chances of hypoparathyroidism and recurrent laryngeal nerve injury (Stosic et al., 2021).

NGS (next-generation sequencing)

By the second half of the last decade, it had become clear that although the number of plausibly actionable genetic alterations was expanding rapidly, no categorical genomics technology existed to identify them all at once. Indeed, as many as three distinct profiling platforms may have been necessary to detect clinically actionable base mutations, copy number alterations, and translocations in a tumor specimen (Ma et al., 2022). However, several powerful DNA sequencing technologies emerged that were radically different from the capillary-based instruments used to analyze the initial reference human genome. These next-generation or massively parallel sequencing (MPS) technologies enabled an unprecedented depth and breadth of genomic interrogation that soon transformed the cancer genome discovery arena (Tarasova et al., 2024).

Next-generation sequencing (NGS) has emerged as a transformative tool in thyroid cancer, offering unprecedented insights into the genetic landscape of tumors. This technology allows for the comprehensive analysis of various genomic alterations, including point mutations, insertions and deletions, copy number variations, gene fusions, and changes in gene expression (Shirali et al., 2024). Such extensive profiling has moved the field beyond the traditional reliance on histopathological features and clinical staging, ushering in an era where biologically informed surgical decision-making is increasingly the standard of care. Therefore, integrating molecular profiling enables personalized surgical planning based on the unique molecular characteristics of each patient’s tumor (Tan et al., 2021).

Within the clinical next-generation sequencing (NGS) platforms, the ThyroSeq panel uniquely integrates comprehensive mutational coverage with predictive diagnostic utility. ThyroSeq screens more than a thousand mutational hotspots in 14 cancer-related genes which encompass major oncogenes and tumor suppressors, including AKT1, BRAF, CTNNB1, GNAS, HRAS, KRAS, NRAS, PIK3CA, PTEN, RET, TP53, TSHR, TERT, and EIF1AX. Furthermore, it identifies 42 gene fusions associated with thyroid cancer including RET, PPARG, NTRK1, NTRK3, ALK, IGF2BP3, BRAF, and MET (Rossi et al., 2022). Beyond fusion and mutation, ThyroSeq provides expression analysis for advanced thyroid-related genes CALCA, PTH, SLC5A5, TG, TTF1, KRT7, KRT20, adding further layers to the multidimensional view of tumor biology. Diagnostic accuracy is further enhanced with the proprietary mitochondrial DNA profiling analysis which has implications in the postoperative management and surgical intervention planning (Steward et al., 2019).

In addressing the problem of managing indeterminate thyroid nodules, tests such as ThyGenX and ThyraMIR can be invaluable. ThyGenX, for instance, is an NGS-based test that screens for certain oncogenic alterations of gene fusions as well as point mutations in BRAF, HRAS, NRAS, KRAS, and PIK3CA, which are significant to thyroid cancer (Wirth et al., 2020). ThyGenX has strong predictive value for certain mutations—detection of BRAFV600E or RET/PTC fusions, is strongly predictive of malignancy, while mutations in RAS and PAX8/PPARγ are associated with variable and less definitive cancer risk. ThyraMIR, on the other hand, is a reflex microRNA test that evaluates the expression of relevant microRNAs such as miR-146, miR-221, and miR-222 to improve risk stratification (Tan et al., 2023). ThyraMIR also provides an additional layer of classification to nodules, marking them as “positive” or “negative” with respect to malignancy risk which refines diagnosis and enhances the performance of molecular testing in the evaluation of indeterminate thyroid nodules (Nikiforov, 2017).

In surgical practice, next-generation sequencing (NGS) has become a critical determinant in deciding the type of approach to be taken in a thyroidectomy. Tumors with low-risk mutations, which include the isolated RAS alterations, are usually viewed as suitable for lobectomy, which is a less extensive and lower-risk surgical option (Prete et al., 2020; Sabi, 2024). On the other hand, high-risk molecular profiles, particularly those with aggressive BRAFV600E or TERT promoter mutations, are more likely to be multifocal, have extrathyroidal extension, and be nodal metastatic. In these cases, total thyroidectomy, with or without central neck dissection, is done to control the disease and to improve the long-term oncologic outcomes (Newfield et al., 2022).

Aside from intraoperative decisions, NGS is equally important for planning adjunct postoperative therapies. A good example is BRAFV600E mutations, which are known to reduce radioactive iodine significantly (RAI) avidity; therefore, making these tumors less responsive to conventional RAI therapy (Paulson et al., 2019; Shirali et al., 2024). In such scenarios, alternative therapies, such as targeted treatments, may yield more favorable results from a clinical standpoint (Khan et al., 2020). Also, RET and NTRK rearrangements are actionable gene fusions that can be targeted with selective kinase inhibitors, particularly in patients with advanced or RAI-refractory disease. In addition, patients with high-risk molecular alterations may be subject to more stringent surveillance because of the greater likelihood of recurrence (Sokilde et al., 2019).

MicroRNA classifiers: expanding molecular precision

Concomitant with mutation analysis based on DNA and RNA sequences, profiling of microRNA (miRNA) has emerged as an essential part in further molecular profiling of thyroid nodules. Researchers designed miRNA profiling with microarray-based platforms due to relatively lower costs. Increasingly, however, they are turning to NGS-based methods due to the increased resolution and dynamic range, which, in many cases, replace or complement microarrays (Yang et al., 2021). Cost-effective miRNA profiling enabled by microarray-based platforms has been on the rise recently. Still, NGS methods have been able to complement and replace these platforms due to their superior resolution and broader dynamic range. Although these changes in technology have been made, the attractiveness of miRNAs as biomarkers remains due to their stability in formalin-fixed tissue and cytological samples, which renders them useful in biopsy materials (Rosignolo et al., 2017).

The clinical relevance of miRNA classifiers in the diagnosis of thyroid cancer has been highlighted in several studies. For instance, evidence was provided for a miRNA assay in indeterminate thyroid nodules on cytology diagnosed as AUS/FLUS and FN/SFN (Santos et al., 2022). Their assay achieved a sensitivity and specificity of 84% along with a negative predictive value (NPV) of 92% which indicates strong assurance in negative results and reduces the chances for unnecessary surgeries. An assumptive framework was constructed in a study involving 109 indeterminate lesions where a set of ten miRNAs was added to a seven-gene mutational panel (Park et al., 2021). Their data provided a positive predictive value (PPV) of 74% and an NPV of 94% which illustrated the benefit of the combined mRNA and microRNA approach to clinical decision making (Santos et al., 2018) (Table 1).

Table 1.

Molecular diagnostics in thyroid cancer: technologies and their impact on surgical planning

Molecular diagnostic tool	Technology type	Targeted mutation/marker	Impact on surgical planning	References
Next-Generation Sequencing (NGS)	DNA sequencing technology	BRAF, RAS, RET, TERT mutations, gene fusions	Facilitates personalized surgery by identifying high-risk mutations for aggressive tumors, leading to more aggressive surgical approaches like total thyroidectomy.	Steward et al., 2019
MicroRNA Profiling (e.g., ThyraMIR)	RNA-based expression analysis	miR-146, miR-221, miR-222	Provides additional risk stratification for indeterminate nodules, potentially reducing unnecessary surgeries for low-risk lesions.	Yang et al., 2021
ThyroSeq Panel	Comprehensive genomic sequencing	AKT1, BRAF, RAS, RET, TP53, TSHR	Allows for comprehensive risk assessment and guides decisions on the extent of surgery (lobectomy vs. total thyroidectomy).	Tan et al., 2023; Sabi, 2024
ThyGenX	Targeted NGS profiling	BRAF, NRAS, HRAS, RET/PTC1, RET/PTC3	Focuses on identifying high-risk genetic alterations for indeterminate thyroid nodules, guiding whether surgical treatment is necessary or active surveillance is an option.	Newfield et al., 2022

The incorporation of miRNA classifiers along with broadened mutation panels. A blinded evaluation of the MPTX (Molecular Profile with ThyraMIR) test on 140 indeterminate thyroid nodules confirmed that not only negative MPTX results but also low-risk MPTX subsets provided a high predictive value for benign histology (94%) (Cabane et al., 2024). Meanwhile, nodules deemed MPTX moderate-risk showed a 53% probability of malignancy, and high-risk MPTX results had a 70% risk of malignancy. These results strengthen the argument in favor of miRNA classifiers as remarkable malignancy predictive markers accessible in lesions that are challenging to discern through cytology examination (Sonmez and Unluturk, 2025).

Significantly, miRNA classifiers have advanced the frontiers of molecular diagnostics from mere identification to directly interfacing with surgical intervention planning. Nodules classed as high-risk based on miRNA expression patterns tend to be more aggressively treated surgically, with total thyroidectomy being performed even in cases considered cytologically indeterminate (Papaioannou et al., 2022). On the other hand, lobectomy or active surveillance is often feasible for nodules exhibiting negative or low-risk miRNA profiles, thereby protecting patients from the dangers linked to excessive surgical procedures (Craig et al., 2023).

ARTIFICIAL INTELLIGENCE AND PRECISION SURGERY IN THYROID CANCER

The integration of artificial intelligence (AI) technologies into the management of thyroid cancer provides important surgical accuracy, which has the potential to change the results of diagnosis and planning of the operation, as well as the surgical decisions, in real time (Bini et al., 2021). The continual development of AI technologies enables surgeons to gather and analyze multiple medical datasets quickly. This advancement transforms their diagnostic and treatment capabilities regarding thyroid tumors (Cece et al., 2025). This chapter studies the contemporary use of AI in treating thyroid cancer by focusing on nodule characterization, risk assessment competencies, surgical navigation, and ethical considerations concerning applying advanced technologies in clinical settings (Taha et al., 2024).

Advanced imaging AI technologies

Advancements in AI technologies have significantly facilitated the field of medical imaging, particularly in imaging and diagnosing thyroid cancers. One of the subdivisions of AI, deep learning (DL), has been exceptionally successful in processing complex biomedical data and in accurately recognizing and extracting relevant features associated with the patient’s condition (Fig. 3). As noted by Habchi and colleagues (Habchi et al., 2023), DL methods are being increasingly incorporated into clinical practice to aid in image evaluation, evaluating the associated risks, and in making clinical decisions. In thyroid cancer diagnosis, DL models have been applied across multiple domains, as detailed below.

Fig. 3.

Flowchart illustrating AI architectures and training methodologies in advanced thyroid cancer imaging. CT: Computed Tomography, MRI: Magnetic Resonance Imaging, AI: Artificial Intelligence, CNNs: Convolutional Neural Networks, CBIR: Content-Based Image Retrieval, RBF: Radial Basis Function, RBMs: Restricted Boltzmann Machines.

AI architectures in imaging-based diagnosis

Convolutional Neural Networks (CNNs): Convolutional Neural Networks (CNNs) stand out as one of the most common deep learning (DL) frameworks employed in medical imaging for the ultrasound scans, computed tomography (CT), and magnetic resonance imaging (MRI) imaging of thyroid glands because they facilitate the learning of features in a hierarchical manner, beginning from the most basic features in raw pixel data (Fig. 3) (Xu et al., 2025).

Typically, the architecture of CNNs includes convolutional layers for feature extraction, pooling layers for dimensionality reduction, and fully connected layers for classification. Besides, these networks are able to detect the spindle-shaped and outlining features of thyroid nodules, which aids in their classification, as well as other complex visual components. Recent research has shown that diagnostic models based on CNNs are much more efficient than models based on traditional image processing, and in some cases, they are as accurate as expert radiologists. Moreover, CNNs allow for the instantaneous and automated interpretation of thyroid cancer scans, enhancing diagnostic consistency and efficiency (Liu et al., 2019).

Using CNNs, Xi et al. (2022) conducted automated detection of thyroid nodules in ultrasound images, retrieving a dataset of 21,532 images from 5,842 patients. Their Model performance was impressive with an AUROC of 98.51% and a sensitivity of 93.5% employing a Region-based CNN (R-CNN) approach. This demonstrates the ability of CNNs to automate the nodule detection process, relieving the clinician’s workload and expediting and streamlining the clinical workflow while maintaining the benchmark accuracy of veteran radiologists.

Kim et al. (2025) applied CNNs to the problem of stratifying the malignancy of thyroid nodules based on ultrasound images. The model achieved a sensitivity of 81.8%, specificity of 86.1%, and accuracy of 85.1%. These results corroborate the capability of CNNs not only to identify and categorize nodules but also to evaluate their malignancy risk. This capability greatly supports triage and other clinical and strategic decisions (Table 2).

Table 2.

Summary of machine learning classifiers applied in medical imaging for diagnostic and risk stratification tasks

Classifier	Data/Modality	Task	Sample size/Dataset	Key performance	References
RBF-SVM	Clinical+Imaging	Distinguish benign vs malignant	1,232 nodules from 724 patients	ML model outperformed human experts; exact metrics in original paper	Zhang et al., 2023b
CNN	Ultrasound images	Thyroid nodule detection	21,532 images from 5,842 patients	AUROC: 98.51%; sensitivity: 93.5% (R-CNN)	Xi et al., 2022
CNN	Ultrasound, radiologist comparison	Malignancy risk stratification	Not specified in snippet	Sensitivity: ~81.8%, specificity: ~86.1%, accuracy: ~85.1%	Kim et al., 2025
CBIR	MRI images (eye/orbit)	Diagnostic accuracy enhancement	48 cases interpreted by 36 radiologists	Accuracy improved from 55.9% to 70.6% with CBIR alone; further to 83.3% combined	Rumberger et al., 2025
RBM	Ultrasound features	Unsupervised feature learning	Various features across studies	RBM listed among feature extraction methods	(Wang et al., 2024

Content-Based Image Retrieval (CBIR): CBIR systems allow the retrieval of clinically analogous images from extensive repositories of medical images, serving as an alternative diagnostic support tool. Unlike older retrieval systems that depend on metadata, CBIR uses the image’s content, including its shapes, texture, and intensity patterns. DL-based CBIR systems, which use CNNs, capture the deep features of a query image and compare them to stored images to retrieve those that are visually similar. This technique aids in the diagnosis of thyroid cancer since it enables a clinician/surgeon to check prior cases with similar morphologic nodules and the relevant diagnostic results (Li et al., 2021).

Rumberger et al. (2025) examined the Use of Content-Based Image Retrieval (CBIR) systems in MRI scans to enhance diagnostic accuracy. The study had 48 cases with 36 radiologist readers. It was found that the application of CBIR enhanced the diagnostic accuracy of thyroid cancer detection. The accuracy improved from 55.9% to 70.6% with the use of CBIR and further improved to 83.3% when used with other methods. These findings demonstrate that the use of CBIR systems enhances confidence in the diagnostics performed. This is especially true when there is a large clinical dataset where radiologists would appreciate a system that retrieves relevant, clinically similar images based on the image content instead of through traditional metadata-based image searches.

Radial Basis Function (RBF): The Radial Basis Function (RBF) kernel is a popular choice in support vector machines (SVMs), a type of machine learning model well-suited for classification tasks involving non-linear data. In thyroid cancer classification, RBF-SVM models are employed to differentiate between benign and malignant nodules by mapping the input data into a higher-dimensional space where linear separation becomes possible (Chen et al., 2024).

Another larger data effort assembled clinicopathological and imaging features from hundreds to over a thousand nodules and compared multiple classical ML classifiers (including SVMs with RBF kernels) against gradient boosting and deep learning baselines. These comparative studies routinely find that carefully engineered feature sets with an RBF-SVM achieve competitive accuracy and are sometimes preferred when dataset sizes are moderate (hundreds to low thousands) and interpretability of features is important. For example, multi-model evaluation that used a cohort of 724 patients / 1,232 nodules directly compared machine learning frameworks (including SVM variants) across tasks such as malignancy detection and presence/absence classification. That work shows classical models like SVM remain relevant when radiomic/clinical features are used. As indicated in Table 2, RBF-SVM models in this dataset outperformed human experts in certain diagnostic tasks (Zhang et al., 2023b).

Restricted Boltzmann Machines (RBMs): Restricted Boltzmann Machines (RBM) are considered to be one of the generative stochastic neural networks used for unsupervised learning purposes. It has a two-layer structure comprising a tangible layer for data input and a hidden layer for capturing the important features. They are essential for feature extraction, pretraining, and deep learning models (Vairale and Shukla, 2020). In the domain of thyroid cancer diagnosis, RBMs can be utilized for high dimensional data such as histopathological imaging and genomic profiling to determine hidden features. These features can be applied for model training or serve as direct support for diagnostic classification. Situations with insufficient labeled data are common. Exploratory data analysis and semi-supervised learning are common in medical research. In these cases, RBMs are very beneficial (Table 2) (Wang et al., 2024).

Training methodologies in imaging AI: The performance of AI models depends not only on their architecture but also on the training strategies used to optimize them (Fig. 3). In thyroid imaging, several techniques are employed:

Transfer learning: In the medical field, Transfer Learning (TL) techniques have been applied to reduce the overfitting challenges associated with insufficient data. Under the TL approach, knowledge sharing and transfer can be done between different tasks. The process consists of two stages: the model is first pretrained on a large dataset, which could be the ImageNet, and later ultra-fine-tuned on the target dataset, which could be the restricted ultrasound images. Put differently, the deep learning (DL) model architecture can be adjusted so that the knowledge gained from one dataset is used on another dataset from a different center (Yazdanpanahi et al., 2024).

Self-supervised learning (SSL): By solving auxiliary tasks such as predicting image rotations or determining missing segments, self-supervised learning (SSL) allows models to derive useful representations from unlabeled data. Furthermore, these representations can be refined with limited labeled data. Such capabilities make SSL appealing in medical imaging, where annotations are costly and time-consuming (Manani and Papavramidis, 2025).

Data augmentation and class imbalance solutions: In response to the problem of sparse data and uneven distribution of classes (i.e., the benign to malignant ratio being greater), augmentation tools like image flipping, rotation, and cropping are implemented. Moreover, performance on minority classes is enhanced by applying re-weighted loss functions, like focal loss, along with synthetic image generation through Generative Adversarial Networks (GANs) (Aljameel, 2022).

Preoperative planning enhanced by AI algorithms

Effective thyroid surgery planning begins well before the first incision, and advancing technologies such as artificial intelligence (AI) are now transforming this critical preparatory phase (Ma et al., 2023). In the past, a considerable portion of surgical planning was reliant on a surgeon’s analysis of imaging studies and their clinical judgment (Lin et al., 2021). While experience continues to hold immense value, AI utilizes pattern recognition algorithms, structural analyses, and predictive modeling that far exceed what a human can do.

AI systems are capable of scrutinizing imaging modalities such as sonography, CT scans, and MRIs with unparalleled accuracy through the use of advanced machine learning algorithms. These systems interpret images at a higher level by performing detail extraction and processing, generating patterns and insights that sometimes surpass the capability of humans, including the most skilled surgeons (Che et al., 2024).

One key benefit of AI in preoperative analysis is the ability to automate the delineation of tumor margin, detection, and assessing the degree of tumor expansion. Modern imaging techniques enhanced through AI guarantee more precise localization of the tumor and accurate estimates of its volume (Jiang et al., 2023). These pieces of information help decide in scenarios like identifying whether a lobectomy or total thyroidectomy is needed so that the balance between having too much surgical aggressiveness and not enough is ensured (Ali et al., 2024).

AI’s contribution to predicting nodal metastasis is equally important. The involvement of lymph nodes with thyroid cancer may, at times, be occult and challenging to reveal with standard imaging techniques (Hernandez-Cruz et al., 2024). AI neural networks developed on extensive databases can detect radiological signs of micrometastases and enhance predictive value (Zhang et al., 2024a). This allows the surgeon to effectively schedule further central or lateral neck dissections in advance and not reactively during surgery as summarized in Table 3.

Table 3.

AI algorithms for surgical planning and intraoperative assistance

AI Technology	Application in Surgery	Benefits	Limitations	References
Image Recognition Algorithms	Analyzes preoperative imaging (CT, MRI, Ultrasound) for tumor localization and margin delineation	Highly accurate tumor detection, better preoperative planning	Requires extensive data and validation across diverse populations	Che et al., 2024
Predictive Models for Lymph Node Metastasis	AI models to predict nodal metastasis based on imaging data	Enhances surgical decision-making for neck dissection	May have limited generalizability without diverse dataset input	Oh et al., 2025; Aweeda et al., 2024
Surgical Navigation Systems	Real-time AI-assisted navigation for identifying critical structures during surgery	Improved intraoperative precision, reduced complications	Requires high-quality imaging and real-time processing, potential technical difficulties	Lixandru-Petre et al., 2025
AI-Driven Postoperative Risk Assessment	Analyzes patient data (clinical, genetic, histopathological) to predict surgical outcomes and recurrence risk	Personalized postoperative care plans, reduces risk of recurrence	Limited by data availability and variability in clinical settings	Gassner et al., 2023

Machine learning-driven real-time surgical navigation: the next frontier

AI in the intraoperative period is one of the most promising technological frontiers for precision in thyroid surgeries. The current research aims to develop machine learning algorithms that process real-time video and imaging data to identify important anatomical structures (Sang et al., 2024). Scientists have created algorithms that correctly detect recurrent laryngeal nerves throughout endoscopic thyroid operations. Early study findings show promising results because measured Dice similarity coefficients established proper functionality when detecting and mapping this significant structure, thus minimizing nerve damage risk (Oh et al., 2025). Such surgical navigation systems, as described in Table 3 and Fig. 4, improve intraoperative precision and reduce complications, but rely heavily on high-quality imaging and real-time processing.

Fig. 4.

Utilizing artificial intelligence, including machine learning technologies, to enhance the diagnosis of thyroid cancer.

AI technology will merge with augmented reality systems using advanced imaging technologies in the forthcoming years to offer additional enhancements for intraoperative navigation (Aweeda et al., 2024). AI-AR systems would supply surgeons with intuitive real-time surgical direction by directly inserting functional and anatomical data into the operating procedure. Combined methods would lead surgeons to perform procedures with better precision, shorter durations, and fewer complications, especially during difficult or complex surgical procedures (Fig. 4) (Nagendra et al., 2023).

Furthermore, AI is adept at delineating human body intricacies, particularly for surgeries like Bilateral Axillo Breast Approach (BABA) thyroidectomies. Unlike open surgeries, the BABA technique must navigate through tighter and more complicated anatomical pathways (Nishiya et al., 2024; Wen et al., 2022). AI assists in identifying and alerting the surgeon about vessels and nerves that could complicate the surgery or make it treacherous. The improvements in the AI Advanced Guiding Systems Overmentioned permissibility bolster the precision for identifying critical structures like the recurrent laryngeal nerve and the parathyroid glands, thus diminishing the risks for inadvertent damage during the surgery (Hu et al., 2022).

Predictive models for surgical outcomes and recurrence risk

The application of AI extends beyond diagnosis to serve as a predictive instrument for tracking postoperative patient results and disease relapse probabilities. Machine learning models accept extensive datasets that include clinical information, histopathological findings, and molecular data to detect hidden correlations that forecast surgical risks and future patient outcomes (Hutchinson et al., 2023). The predictive XGBoost algorithm exhibits high discriminative ability by achieving an area under the receiver operating characteristic curve (AUROC) values up to 0.88 in thyroid cancer recurrence predictions (Lixandru-Petre et al., 2025).

The development of AI-driven risk models now prioritizes explainable aspects in their creation processes. Predictive models can be interpreted using Local Interpretable Model-agnostic Explanations (LIME) and Morris Sensitivity Analysis, which identify the contribution of each factor to the model’s predictions (Wu et al., 2023). Interpretability is essential to generating clinical trust and enabling active participation in medical decision discussions between physicians and patients. Thyroid cancer treatment benefits from AI tools that deliver customized risk assessments, thus helping physicians plan surgeries better and create patient-led decisions during care, as summarized in Table 3 (Gassner et al., 2023).

Ethical considerations in AI-assisted decision-making

Surgical practitioners must understand and resolve the ethical, legal, and social concerns that emerge due to the rising presence of AI in surgery. Transparency and explainability represent the most critical elements in this context. Doctors must decode and comprehend AI output recommendations to preserve their medical accountability alongside patient care security (Choi et al., 2017). The importance of medical responsibility emerges when considering this topic. Concerning human interactions with clinicians, there are defined liability limits regarding how AI systems interact with the patient about a judgment input concerning a negative result (Fernández Velasco et al., 2024).

The problems identified in healthcare AI extend beyond fairness to include bias. When non-diverse data is used to train AI systems, they can perpetuate existing inequities in healthcare (David et al., 2024). Without active intervention, these algorithms will be systematically underperforming for various groups of patients. To achieve machine-level success, exhaustive algorithmic testing within diverse population strata and inclusive information preparation for the machine are prerequisites. Information given to patients regarding their consent needs careful examination (Table 3). Clinicians are ethically obligated to disclose the use of AI tools in clinical decision-making, ensuring that patients understand both the capabilities and limitations of these technologies (Lin et al., 2021).

CHALLENGES AND FUTURE DIRECTION

The future of surgical approaches to thyroid cancer incorporates new diagnostic and treatment technologies into streamlined, patient-friendly care pathways. A key component is the creation of integrated diagnostic systems that include radiological imaging, molecular profiling, and AI analytics (Toraih et al., 2021). The integration will facilitate greater adaptability and personalization of risk assessments to ensure that surgical plans are biologically intelligent, appropriately constructed, and aligned with tumor biology. Simultaneously, the predominance of robotic-assisted minimally invasive techniques like the bilateral axillary-breast approach (BABA) must be balanced (Jung et al., 2022). Increased surgical precision, decreased operative duration, broadened indications for complex and advanced malignancies, and greater diagnostic scope will drive procedural evolution (Durante et al., 2023).

However, applying these techniques should be accompanied by a greater focus on global, equitable access. Standardized training development and subsidized robotic equipment in lower- and middle-income countries will help address the pronounced disparities in thyroid cancer care. Furthermore, the validation of molecular diagnostics and AI-inspired models will need long-term, multicenter studies to assess recurrence-free survival, complication-free survival, quality of life, and economic value alongside robust clinical and prognostic relevance.

The ethical and regulatory concerns accompanying artificial intelligence in the clinic AI must be integrated into clinical decision systems, necessitating more thoughtful attention. It should deal with issues like bias in algorithms, inclusively retraining datasets, interfacing with equity and accountability frameworks in care provision, and informed consent. All highlighted focal concerns attest to the commitment from different areas towards refining precision, conscientious inclusivity, and ethical frameworks guiding thyroid cancer surgery.

CONCLUSION

The management of thyroid cancer is undergoing important changes due to the integration of surgery, artificial intelligence (AI), and molecular diagnostics. BABA robotic thyroidectomies resolve the issues of open surgery regarding cosmetic and intraoperative finesse accuracy. Further, profiled molecular evaluation by next-generation sequencing and microRNA classifiers provides biotechnological intel for preoperative assessments, allowing surgeons to decide on the scope of surgery and node dissection. Moreover, AI microscope integration shifts the paradigm of supplemental radiology, navigation, and intraoperative histology, bringing automation to support technologies.

These disparate developments provide strong evidence for moving towards precision endocrine surgery. Such a surgical approach incorporates elastic parameters including, but not limited to, patient-specific markers alongside tumor size and location to make informed decisions on tumor aggressiveness, tumor behavior, or if surgery is strategically beneficial for the patient. While integrating various technologies into one domain provides promising results, some unique challenges are barriers that require immediate attention. Addressing those challenges will make the use of radiology, AI technologies, and machine learning more effective: the challenges include universal accessibility, assessing the reliability of long-term effects, and establishing ethical boundaries.

Thyroid surgery in the upcoming years is expected to add new dimensions of safety, efficiency, and alignment with patient-centric considerations like personal aesthetics concordant with women’s interests and preferences. A multidisciplinary approach integrating surgical techniques with molecular biology and computation will transform the future of thyroid cancer to enable a paradigm shift, aiming not only at fulfilling survival goals but also refining the quality and enabling the patients to prosper vitally and holistically, achieving comprehensive thyroid health.