Thermographic differentiation of thyroid carcinoma and adenoma in indeterminate cytology nodules: insights from intra- and preoperative imaging

Abstract

Background

Thyroid carcinoma (TC) is the most common endocrine system tumor, whose diagnosis is performed through the fine-needle aspiration biopsy (FNAB), a minimally invasive procedure that uses a thin needle to extract cells from suspicious thyroid nodules for cytological examination. However, in 15–25% of cases there is no certain preoperative response (indeterminate lesions classified as TIR 3A or TIR 3B). In this perspective, developing alternative diagnostic methods is highly desirable. To this aim, infrared thermography (IRT) could be a valuable tool to detect the presence of thyroid cancer. The objective of this study is to investigate the feasibility and the effectiveness of employing IRT to discriminate thyroid cancers in clinical practice.

Methods

The study included 25 participants who underwent to total thyroidectomy surgery: 6 with TC, 5 with thyroid adenoma (TA), and 14 with thyroid goiter (TS). IRT recordings were conducted twice for each patient: first after anesthetic induction with the cervical region exposed and second after intraoperative exposure of the thyroid gland. Two regions of interest (ROIs), corresponding to the right and left thyroid lobes, were analyzed for maximum (Tmax), minimum (Tmin), and average (Tavg) temperatures. The difference between the right and left thyroid lobes sides of the three evaluated metrics was calculated and assessed through t-tests for independent samples. Additionally, the receiver operating characteristic (ROC) curves were calculated to define a threshold for the difference in the considered metrics that distinguishes between subjects with carcinoma, adenoma, and goiter. Finally, a correlation analysis between the inner and external temperature was performed for the three groups.

Results

For the cervical region measurements, Tmax successfully distinguished between carcinoma and adenoma (P=0.035) and between goiter and carcinoma (P=0.034). Meanwhile, Tavg showed a significant difference between goiter and carcinoma (P=0.017). For measurements taken on the exposed gland, Tmax could differentiate between carcinoma and adenoma (P=0.044) and between goiter and carcinoma (P=3.81×10⁻⁴). Tavg was significant in distinguishing goiter from carcinoma (P=0.004) and goiter from adenoma (P=0.028).

Conclusions

This pilot study determined the effectiveness of IRT in identifying thyroid nodular formations suspected of being neoplastic, providing specialists and patients with an additional diagnostic tool for managing thyroid nodules, especially those cytologically characterized as indeterminate risk neoplastic lesions. IRT could represent a valid complementary tool that may assist the physician in the diagnostic process of thyroid nodules.

Introduction

Thyroid cancer is the prevailing neoplasm of the endocrine system, and its frequency has significantly risen in recent years (1,2), with the most recent estimates indicating approximately 567,233 new cases diagnosed annually worldwide (3,4). Indeed, this epidemiological landscape underscores the importance of continuous monitoring and refined screening practices for early tumor diagnosis. The classification of thyroid cancer is currently performed in agreement with the 5th Edition of the World Health Organization (WHO) Classification of Endocrine and Neuroendocrine Tumours (5). This classification emphasizes both histological features and molecular characteristics, leading to a more specific categorization that aids in the diagnosis and treatment of thyroid tumors (6,7).

Follicular neoplasms of the thyroid gland may be categorized as either benign adenoma or malignant carcinoma (8). Timely detection of malignant neoplasms enhances the probability of successful therapy and elevates the overall survival rate of individuals with cancer (9).

Thyroid cancer diagnosis involves a multi-faceted approach, utilizing both imaging techniques and histopathological evaluations. Ultrasound imaging is the primary method employed for thyroid nodule evaluation and offers detailed morphological assessment. Specifically, grayscale ultrasound is usually employed for initial assessments, providing information regarding nodule size and characteristics that may indicate malignancy. Additionally, color Doppler ultrasound imaging allows for the identification of features particularly indicative of thyroid carcinoma (TC), such as increased vascularity (10,11). While ultrasound remains the first-line imaging modality, computed tomography (CT) and magnetic resonance imaging (MRI) can provide additional insights, particularly in evaluating lymph node metastases, mediastinal invasion, or in cases where other imaging is inconclusive. However, the role of CT and MRI remains more limited compared to ultrasound due to cost and exposure concerns (12,13). In fact, fine-needle aspiration biopsy (FNAB) still remains the gold standard technique for distinguishing between benign and malignant thyroid nodules (14). Other tools as molecular tests capable to analyze genetic alterations (e.g., BRAF mutations, RAS, RET/PTC or PAX8/PPARγ rearrangements) or gene expression profiles are also employed to distinguish benign from malignant lesions. Tests such as ThyroSeq, Afirma GSC and ThyGeNEXT/ThyraMIR are among the most widely used. For example, ThyroSeq evaluates a panel of genes associated with thyroid cancer, demonstrating high sensitivity (90–94%) and specificity (92–96%) in detecting malignancy (15).

According to the 2015 American Thyroid Association guidelines, molecular tests are recommended in indeterminate cases to guide the choice between observation and surgery (16). A study by Alexander and Cibas [2022] found that BRAF and RAS analysis reduced unnecessary surgery by 30–50% (17). However, limitations include high costs, limited availability and the need to correlate results with the clinical context.

Moreover, it was demonstrated that the integration of Next-Generation Sequencing techniques improves the identification of rare variants, reducing false negatives (18). Despite this, standardization of protocols and interpretation of results remain open challenges.

Although the integrated use of cytology and molecular biology represents a significant advance in the personalized management of thyroid nodules, balancing diagnostic efficacy and therapeutic appropriateness, costs remain high and not all centres have these means available.

Importantly, The Bethesda System for Reporting Thyroid Cytopathology (TBSRTC) and the European Thyroid Imaging Reporting and Data System (EU-TIRADS) are two classification systems used for the evaluation of thyroid nodules, each focusing on different aspects. The TBSRTC classification is a cytological system, based on the microscopic examination of cells obtained through FNAB. In contrast, the EU-TIRADS classification is an ultrasound-based system that evaluates nodules according to their imaging characteristics.

In Italy, the results of the FNAB are usually interpreted using the Italian thyroid cytology classification system composed by five diagnostic cytology categories (TIR 1 to TIR 5), which is comparable with the TBSRTC classification system (19,20). In detail, the TIR system categorizes FNAB cytology (FNAC) results into six classes: TIR 1, non-diagnostic; TIR 2, benign; TIR 3A, low-risk indeterminate lesion; TIR 3B, high-risk indeterminate lesion or suspected follicular neoplasm; TIR 4, suspicious for malignancy; and TIR 5, malignant. Particular attention is given to TIR 3A and TIR 3B, which correspond to Bethesda’s Thy3a and Thy3f categories (21). Both TIR 3A and TIR 3B represent indeterminate cytological findings but differ significantly in their associated risk of malignancy and subsequent clinical management. TIR 3A corresponds to atypical or hyperplastic lesions with minimal cytological atypia and is associated with a relatively low risk of malignancy (approximately <10%). Conversely, TIR 3B includes lesions with a higher degree of architectural and cytological atypia, suggestive of follicular neoplasia, with an estimated malignancy risk between 15% and 30%. Whereas TIR 3A cases are often managed conservatively with repeat FNAC and clinical surveillance, TIR 3B usually requires surgical intervention, typically lobectomy, for definitive diagnosis. The distinction between TIR 3A and TIR 3B thus plays a crucial role in optimizing patient management and avoiding unnecessary surgery while ensuring timely treatment of potentially malignant nodules.

In the literature, it is well known that thyroid cancer cells secrete cytokines to stimulate angiogenesis, leading to increased vascularization and distorted vessel distribution (22). Moreover, the endothelin axis has been associated with cancer vascularization, exhibiting altered expression across several cancer types, including papillary TC (23). Notably, infrared thermography (IRT) is a contactless imaging technique able to detect skin temperature variations, which are strictly related to the microcirculation modulations (24). Thus, given the anatomical vascularization changes and the lymphangiogenesis associated with the tumor, IRT could be successfully employed for thyroid cancer assessment. In fact, the use of IRT for the detection of thyroid cancer has gained attention in recent years due to its potential as a non-invasive diagnostic tool. For instance, Chantasartrassamee et al. demonstrated that artificial intelligence (AI) algorithms applied to IRT data acquired during a cooling procedure can be effective to identify thyroid lesions with respect to surrounding healthy tissue (25). Supporting these findings, Bahramian and Mojra discovered that, after a neck cooling procedure, the temperature in the area directly in front of the malignant tumor increased by 1–1.5 ℃ compared to the surrounding healthy tissue (26). Similarly, Damião et al. showed that thermographic imaging could aid in identifying nodules with histological diagnoses of malignancy, suggesting that temperature differences exist between malignant and benign nodules (27). In a comparative study, Alves and Gabarra found that IRT might detect changes in tissue temperature earlier than conventional imaging techniques, potentially improving early detection when selecting nodules for FNAB (28). Moreover, by focusing IRT on the neck region, it was demonstrated that the tumor tissue had a greater level of metabolic activity compared to the healthy tissue (29). In line with these findings, Flores et al. further highlighted that IRT could detect the increased metabolic activity characteristic of malignant thyroid tissues through the employment of artificial neural networks (30). In addition, Bahramian and Mojra also implemented an artificial neural network to evaluate the connections between the thermo-physical characteristics of the thyroid and the temperature distribution on the skin of the neck (31). Furthermore, Damião et al. emphasized the value of thermal monitoring as an auxiliary parameter in thyroid nodule assessment, evaluating the association between thermographic anomalies and tumor malignancy employing convolutional neural networks (32). Notably, most of the studies employing IRT for detecting thyroid nodularity investigate only temperature patterns of the skin of the neck. However, it should be highlighted the skin temperature modifications may reflect the anatomical changes of the thyroid lodge due to the pathology. Hence, investigating directly the temperature of the gland during surgery may support this hypothesis.

The objectives of this study are to evaluate the effectiveness of IRT to discriminate TC, TA, and normal struma contained in the thyroid goiter (TS) from features extracted from the skin temperature distribution of the frontal region of the neck and from the inner gland temperature, and secondarily to find a correlation between the surface and inner temperature of the thyroid lodge. A particular focus was placed on the capability of IRT recorded on the neck’s skin to identify TIR 3A and TIR 3B cases. We present this article in accordance with the STARD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1482/rc).

Methods

Subjects

The patients included in the study were recruited between October 2024 and January 2025 at the General Surgery Department of the Hospital Casa di Cura Pierangeli in Pescara. In this study, IRT images and temperatures measurements were prospectively collected from a total of 30 patients who underwent to total thyroidectomy surgery. The inclusion criteria were major age, patients not undergoing radiotherapy treatments and FNAB of highly suspicious thyroid nodules at ultrasound examination (EU-TIRADS 3). The exclusion criteria were previous neck surgeries, hyperthyroidism, chronic intake of anti-inflammatory drugs (cortisone or immunosuppressants) or anti-thyroid drugs (methimazole or propylthiouracil), minor age and issues related to thermographic data acquisition. Regarding the IRT measurements, 5 participants were excluded because of technical reasons, none of which is related to clinical characteristics of the patients: obstruction of the field of view by surgical drapes, personnel and the anaesthesia circuit during acquisition (n=2 cases), impossibility of completing the acclimatization period or the acquisition protocol because of operating-room scheduling constraints (n=2 cases) and incorrect framing or focusing of the cervical region, so that one of the two ROIs could not be defined (n=1 case). Since these causes are purely technical or logistic and occurred before any histological information was available, the exclusion process cannot have introduced a selection bias with respect to the study outcome.

Based on these criteria, we selected 25 data samples from the initial 30 available. For each data sample collected from patient, clinical and imaging evaluation were performed before surgery.

Clinical-anamnestic data showed that our series consisted of young and adult patients, the participants’ ages ranged from 30 to 79 years, with an average age of approximately 56.24 years. The patients recruited were respectively 16 female and 9 male. All 25 patients included in the study underwent a pre-operative ultrasound study of the thyroid, 13 also underwent a non-contrast neck-chest CT scan, and 3 underwent thyroid scintigraphy with I-131. Thirteen out of 25 patients underwent FNAB with cytological examination. It is wort to mention that all FNAB procedures were performed at least four months before surgical interventions in order to not alter the original physiological temperature of the tissue at the time of temperatures recording with IRT. Among the patients who had a thyroid ultrasound, one received EU-TIRADS 2 evaluation, three EU-TIRADS 5, and two EU-TIRADS 4. All three patients who underwent scintigraphy had hypofunctioning areas corresponding to ultrasonographically identified nodules. Of the patients who underwent FNAB, one patient was diagnosed with a TIR 5 nodule; three patients were diagnosed with TIR 4 nodules, one of which had previously been reported as TIR 2; three patients were diagnosed with TIR 1 nodules (one of which was previously reported as TIR 2), and in one of these patients, the cytological examination was performed on two distinct nodules, one of which was TIR 2. There were six indeterminate cytology diagnoses: three patients were diagnosed with TIR 3A nodules (one of which was previously TIR 2) and in one of these three patients, the cytological examination was performed on two distinct nodules; three patients were diagnosed with TIR 3B nodules (one of which was previously TIR1). Overall, 4 patients repeated FNAC (TIR 4 vs. TIR 2, TIR 3B vs. TIR 1, TIR 1 vs. TIR 2, TIR 3A vs. TIR 2). All patients enrolled in the study underwent pre-operative evaluation, including blood tests, electrocardiogram, and anesthesiological assessment. After providing informed consent, all patients underwent Total Thyroidectomy surgery through a cervicotomy under general anesthesia, with the specimen sent for definitive histological examination.

IRT instrumentation and data acquisition

Patients were instructed to avoid consuming vasomotor substances such as tea, coffee, and tobacco within 24 hours before the surgery and to avoid using moisturizers on the neck region. Recordings were performed in the operating room, in an adequately sized environment, at least 15 minutes after the subject arrived in the operating room, at a constant temperature of 21 ℃, with a humidity rate of 55%, avoiding direct ventilation and lighting on the region of interest. The recordings were conducted with a FLIR SC660 digital thermal camera (FLIR, Wilsonville, OR, USA), with an FPA 640×480 bolometer, and sensitivity <30 mK at 30 ℃. The camera was positioned approximately 60 cm from the subject’s neck, with its optical axis perpendicular to the imaged surface (orthogonal to the anterior cervical plane for the pre-operative series and to the plane of the exposed gland for the intra-operative series) and aligned with the midline of the region of interest. The sampling frequency was 10 Hz. For both series, the camera object parameters were set to an emissivity of 0.98, a reflected apparent temperature of 21 ℃, an object distance of 0.6 m and a relative humidity of 55%; atmospheric attenuation was negligible at this working distance. The value of 0.98 is the established emissivity of human skin in the long-wave infrared band and was retained for the intra-operative series because the infrared emissivity of moist, blood-perfused biological tissue is dominated by its water content and lies in the same range (0.96–0.99). Two series of recordings were made for each patient. The first series was conducted immediately after the completion of anesthetic induction, with the cervical region exposed after the placement of retro-scapular support, while the second series was recorded after the intraoperative exposure of the gland, externalized by the pre-thyroid muscles. Notably, IRT acquisitions were performed exclusively on the day of surgery and was never carried out during or immediately after FNAB.

Statistical analysis

The sample used for the statistical analysis consisted of 25 participants, including 6 TC, 5 TA, and 14 TS. Two ROIs were considered from the thermal images: the right and left sides of the thyroid, both for the images acquired before the operation and for the images acquired with the thyroid exposed (Figure 1). For the pre-operative images, each ROI was positioned over the cutaneous projection of the corresponding lobe, delimited medially by the tracheal midline, laterally by the medial border of the sternocleidomastoid muscle, inferiorly by the jugular notch and superiorly by the cricoid cartilage. For the intra-operative images, each ROI was positioned over the exposed surface of the corresponding lobe, excluding retractors, surgical instruments and gauzes. ROIs were placed manually by a single trained operator, blinded to the histological diagnosis, using a custom-made software for IRT video tracking. The elliptical shape was determined by the ROI tools available in the software, and the axes were set so that the ROI sampled a representative portion of the lobar region while remaining within the anatomical boundaries described above. Within each subject, the two ROIs were given identical dimensions and mirrored positions with respect to the midline, in order to sample the two sides equivalently. The ROI was intentionally defined as a landmark-based lobar region rather than as a segmentation of the nodule, consistently with the intended use of the technique as a fast-screening tool, in which the nodule location is not known a priori.

Figure 1.

Representative pre-operative and intra-operative thermograms. The red elliptical outlines indicate the two ROIs from which the maximum, minimum and average temperatures were extracted. In the pre-operative image, the ROIs enclose the cutaneous projection of the right and left thyroid lobes on the anterior lower neck, whereas, in the intra-operative image, they enclose the surface of the corresponding exposed lobes after externalization through the pre-thyroid muscles. The elliptical shape is imposed by the ROI tools of the analysis software. The colour bar reports temperature in ℃. ROI, region of interest.

From these ROIs, three metrics were extracted: maximum temperature (Tmax), minimum temperature (Tmin), and average temperature (Tavg).

Based on the hypothesis that the tumor causes a change in the overlying skin temperature, the absolute value of the difference between the right and left thyroid lobes sides of the three evaluated metrics was calculated. The difference between the three groups was assessed through t-tests for independent samples. Additionally, to evaluate the ability to identify the presence of pathology in individual participants, the receiver operating characteristic (ROC) curve and the confusion matrix (CM) were calculated. This allows the definition of a threshold for the difference in the considered metrics that distinguishes between subjects with carcinoma, adenoma, and goiter. Furthermore, it enables the assessment of the method’s performance in terms of sensitivity and specificity. It is important to note that the ROC curves were evaluated by considering one class (TC, TA, and TS) against all others (TC vs. TA + TS, TA vs. TC + TS, TS vs. TC + TA). In addition, the capability of the approach to identify TIR 3A and 3B cases was verified. Finally, a correlation analysis (i.e., Pearson correlation coefficient) between the inner and external temperature was performed for the three groups to assess the capability of superficial IRT to assess modifications of the metabolism in the thyroid region associated with the pathology. It is relevant to confirmed that the operator who positioned the ROIs and extracted the temperature metrics was blinded to the histopathological diagnosis, which were disclosed after all thermal metrics had been extracted and stored.

Ethical consideration

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of the “G. d’Annunzio” University of Chieti-Pescara, with approval covering the research activities conducted at the Casa di Cura Pierangeli Hospital in Pescara (approval No. Prot. 392490/24, 09/10/2024 - C.Et.R.A. Pescara). All patients gave their informed consent prior to their inclusion in the study.

Results

Canonical diagnostic tools results

In the 25 patients enrolled in the study, the final histological examination identified 6 patients with TC (24%), 5 patients with TA (20%), and 14 patients with TS (56%).

Among the patients with carcinoma, one out of six had a triple synchronous tumor localization (16% of 6; 4% of 25), while another had bilateral localization of papillary carcinoma (16% of 6; 4% of 25). In one out of six patients, an adenoma associated with a microcarcinoma on the same side was found (16% of 6; 4% of 25), while the remaining three patients had single localizations (50% of 6; 12% of 25). In total, the histological examination of the specimens found 9 malignant neoplasms, 8 carcinomas and 1 microcarcinoma. Considering the distribution of TIR 3A and TIR 3B across the TC, TA and TS, results that two histological diagnoses of TC were previous defined as TIR 3B, two TA were defined as TIR 3B, and three TS were defined as TIR 3A.

IRT results

The one-way analysis of variance (ANOVA) delivered not statistically significant differences between Tmax across the three groups for both the thyroid and skin Tmax (F=0.15, P=0.859; F=0.39, P=0.681, respectively). Also, regarding Tmin, no significant differences were assessed for both the thyroid and skin Tmin (F=0.05, P=0.955, F=2.26, P=0.126). The same finding was assessed for Tavg (F=0.29, P=0.749; F=1.83, P=0.182). Although the ANOVA comparisons were not significant, t-tests assessing differences between the different groups separately were performed. The results of t-tests comparing the right and left side of the neck conducted among the TC, TA, and TS groups for the metrics evaluated from pre-operative image are reported in (Table 1).

Table 1. Results of the t-tests conducted for the thermal metrics evaluated on the right and left side of the neck skin temperature among the TC, TA, and TS groups.

Metric	Comparison	T-test	P value	Adjusted P value (FDR correction)
Tmax	TC vs. TA	2.465	0.035	0.090
	TS vs. TC	−2.267	0.034	0.090
	TS vs. TA	−0.454	0.654	0.654
Tmin	TC vs. TA	−0.926	0.385	0.462
	TS vs. TC	−1.608	0.122	0.200
	TS vs. TA	−1.476	0.155	0.233
Tavg	TC vs. TA	0.759	0.472	0.500
	TS vs. TC	−2.587	0.017	0.077
	TS vs. TA	−1.814	0.085	0.153

FDR, false discovery rate; TA, thyroid adenoma; Tavg, average temperature; TC, thyroid carcinoma; Tmax, maximum temperature; Tmin, minimum temperature; TS, thyroid goiter.

The data obtained from Tmax recordings showed significantly and evident difference between the left and right sides thyroid lobe in TC, TA and TS. The comparative analysis of Tmax recordings allowed to state that is significantly possible to discriminate between TC and TA, and also between TS and TC based on their Tmax. Meanwhile, Tmax comparison between TS and TA in not significant. Tmin is not significant for any of the comparisons made, whereas the Tavg, however, significantly distinguishes only between TS and TC. The results of the t-tests conducted for the thermal metrics evaluated on the exposed thyroid are reported in (Table 2).

Table 2. Results of the t-tests conducted for the thermal metrics evaluated on the left and right sides of the exposed thyroid among the TC, TA, and TS groups.

Metric	Comparison	T-test	P value	Adjusted P value (FDR correction)
Tmax	TC vs. TA	2.335	0.044	0.099
	TS vs. TC	−4.306	3.81×10−4	0.003
	TS vs. TA	−1.257	0.225	0.312
Tmin	TC vs. TA	−0.926	0.385	0.462
	TS vs. TC	−2.026	0.056	0.112
	TS vs. TA	−5,455	2.43×10−5	<0.001
Tavg	TC vs. TA	0.759	0.472	0.500
	TS vs. TC	−3.293	0.004	0.024
	TS vs. TA	−2.356	0.028	0.090

FDR, false discovery rate; TA, thyroid adenoma; Tavg, average temperature; TC, thyroid carcinoma; Tmax, maximum temperature; Tmin, minimum temperature; TS thyroid goiter.

From the (Table 2), it is evident that the difference in Tmax between the left and right thyroid lobes sides can significantly discriminate TC from TA and TS from TC, but it cannot discriminate between TS and TA. Tmin, on the other side, can distinguish between TS and TA but is not significant for the other comparisons considered. Tavg distinguishes between TS and TC, TS and TA, but not TC and TA. The temperature metrics evaluated were used to compute the ROC curve associated with the ability to discriminate between TC, TA, and TS. The sensitivity, specificity, accuracy, and threshold associated with the three metrics considered in the two conditions are reported in (Table 3).

Table 3. Results associated with the ROC curve analysis expressed as sensitivity, specificity, accuracy, and threshold for the three metrics considered in the two conditions.

Condition	Metric	Comparison	Sensitivity (95% CI)	Specificity (95% CI)	Accuracy (95% CI)	Threshold (℃)
Neck	Tmax	TC vs. TA + TS	66.7 (30.0–90.3)	84.2 (62.4–94.5)	75.5 (54.9–96.0)	>0.673
		TA vs. TS + TC	60.0 (23.1–88.2)	55.5 (34.2–74.2)	57.7 (33.4–81.6)	<0.498
		TS vs. TA + TC	62.7 (38.8–83.7)	92.7 (62.3–98.4)	77.7 (62.4–92.8)	<0.645
	Tmin	TC vs. TA + TS	66.7 (30.0–90.3)	57.9 (36.3–76.9)	62.3 (40.4–84.2)	>0.259
		TA vs. TS + TC	60.0 (23.1–88.2)	40.0 (21.9–61.3)	50.0 (26.0–74.0)	<0.259
		TS vs. TA + TC	57.1 (32.6–78.6)	72.7 (43.4–90.3)	64.9 (46.5–83.4)	<0.284
	Tavg	TC vs. TA + TS	83.3 (43.6–97.0)	63.2 (41.0–80.0)	73.2 (54.8–91.7)	>0.340
		TA vs. TS + TC	60.0 (23.1–88.2)	65.0 (58.4–74.9)	62.5 (46.8–73.2)	>0.597
		TS vs. TA + TC	64.3 (38.8–97.0)	90.9 (75.9–94.9)	77.6 (57.7–94.1)	<0.430
Thyroid exposed	Tmax	TC vs. TA + TS	83.3 (58.4–91.9)	94.7 (62.5–97.0)	89.0 (62.6–94.1)	>0.696
		TA vs. TS + TC	60.0 (23.1–88.2)	80.0 (58.4–91.9)	70.0 (46.8–93.2)	<0.339
		TS vs. TA + TC	64.3 (38.8–83.7)	81.8 (52.3–94.9)	73.1 (56.1–90.0)	<0.449
	Tmin	TC vs. TA + TS	83.3 (43.6–97.0)	68.4 (46.0–84.6)	75.9 (57.7–94.1)	>0.696
		TA vs. TS + TC	40.0 (31.8–76.9)	75.0 (53.1–88.8)	57.5 (34.0–81.0)	<0.267
		TS vs. TA + TC	100 (78.5–100.0)	54.5 (28.0–78.7)	77.3 (62.6–92.0)	<1.403
	Tavg	TC vs. TA + TS	66.7 (33.0–90.3)	63.2 (41.0–80.9)	64.9 (43.2–86.7)	>0.466
		TA vs. TS + TC	40.0 (11.8–76.9)	65.0 (43.3–81.9)	52.5 (28.6–76.4)	<0.187
		TS vs. TA + TC	71.4 (41.5–88.3)	81.8 (52.3–94.9)	76.6 (60.2–93.1)	<0.341

CI, confidence interval; ROC, receiver operating characteristic; TA, thyroid adenoma; Tavg, average temperature; TC, thyroid carcinoma; Tmax, maximum temperature; Tmin, minimum temperature; TS thyroid goiter.

One TIR 3A is correctly identified as struma by Tmax and Tavg of the neck, but it is not correctly identified by Tmin. Another patient categorized as TIR3B is correctly identified as adenoma by Tmax and Tmin, but it is classified as struma by Tavg. A participant classified as TIR 3A was correctly identified as struma by Tmax, but it is not properly categorized by the other two metrics. Two patients classified as TIR 3A was properly identified by all the metrics as struma. However, one participant with TIR 3B was wrongly evaluated as carcinoma instead of adenoma by all the metrics. The results of the Pearson correlation between Tmax, Tmin and Tavg of the neck and of the exposed thyroid are presented in (Table 4), whereas (Figure 2) shows the scatter plot between them.

Table 4. Results of the Pearson correlation among the Tmax, Tmin, and Tavg of the neck and of the exposed thyroid.

Group	Metric	Correlation coefficient	P value
Carcinoma	Tmax	−0.282	0.430
	Tmin	−0.707	0.022
	Tavg	−0.462	0.179
Adenoma	Tmax	0.774	0.024
	Tmin	0.610	0.109
	Tavg	0.252	0.548
Struma	Tmax	0.266	0.117
	Tmin	0.123	0.476
	Tavg	0.174	0.311

Tavg, average temperature; Tmax, maximum temperature; Tmin, minimum temperature.

Figure 2.

Scatter plot among the Tmax, Tmin, and Tavg of the neck and of the exposed thyroid. Tavg, average temperature; Tmax, maximum temperature; Tmin, minimum temperature.

Discussion

This study investigates the feasibility of employing IRT to discriminate thyroid cancers. Due to the sample size limitations, the current data we presented and the possible impact on clinical practice we discussed in our study have an explorative scope. Specifically, in this pilot study the first investigation was focused on the temperature difference measured with IRT between the two sides of the skin covering the thyroid and the temperature difference between the left and right lobes of the exposed thyroid. In detail, Tmax, Tmin and Tavg temperatures of each ROIs were evaluated. Particularly, this study explores the capability of employing IRT to provide a reliable diagnosis in those indeterminate cases represented by the TIR 3A and 3B of the TIR classification system of the FNAC results. Importantly, under a clinical point of view, these two categories represent a condition of diagnostic uncertainty. Hence, having access to a technology capable of identifying the tumor pathology even in TIR 3A and TIR 3B cases could have a significant impact on clinical practice. In fact, these categories often lead to unnecessary surgical interventions, thus developing a method able to differentiate between benign and malignant lesions in these indeterminate cases could reduce the reliance on invasive or inconclusive procedures. Moreover, it is worth highlighting that as a non-invasive, radiation-free, and cost-effective technique, IRT offers a safe and accessible auxiliary method to conventional diagnostic modalities. The findings of this study demonstrated that, for the preoperative skin recordings, Tmax significantly distinguished carcinoma from adenoma (P=0.035) and goiter (P=0.034), Tavg differentiated carcinoma from goiter (P=0.017), while Tmin showed no significant differences. In the intraoperative gland recordings, Tmax discriminated carcinoma from adenoma (P=0.044) and from goiter (P=3.81×10⁻⁴). Tavg was effective in distinguishing goiter from both carcinoma (P=0.004) and adenoma (P=0.028), whereas Tmin significantly differentiated goiter from adenoma (P=2.43×10⁻⁵), but not carcinoma from adenoma. The ROC curve-based analysis revealed that Tmax evaluated for the exposed gland achieved the best performance in detecting carcinoma vs. non-carcinoma cases, with a sensitivity of 83.3%, specificity of 94.7%, and overall accuracy of 89% at a threshold of >0.696 ℃ in side-to-side temperature difference. Notably, in order to evaluate the capability of IRT to distinguish between carcinoma, adenoma, and goiter, particularly for the cases of indeterminate diagnosis, in order to avoid unnecessary surgeries, it is necessary to evaluate the performance obtained when measuring the neck temperature rather than the one reached with the exposed thyroid. Importantly, IRT demonstrated potential in identifying TIR 3A and TIR 3B lesions when considering the difference of the temperature of the two sides of the neck, corroborating the hypothesis that thermal anomalies correlate with the malignant or benign histology, suggesting IRT could serve as a supportive tool in preoperative risk stratification.

Another important finding of this study is represented by the correlation between internal (intra-operative) and superficial (pre-operative) temperature measurements. In fact, if superficial skin temperature reliably reflects the underlying thermal behavior of the thyroid gland, IRT could be employed as a non-invasive tool to infer metabolic or vascular anomalies associated with malignancy. Particularly, a significant positive correlation between the external and inner temperature was assessed for the adenoma cases, a weak positive correlation for the struma, and a strong negative correlation for the carcinomas. Given the relatively small sample size, we cannot exclude that statistical instability and sampling variability may have contributed to the observed differences. For these reasons, these findings should be considered exploratory and require confirmation in larger cohorts.

These findings suggest different metabolic and circulatory patterns in malignancies with respect to benign cases.

In detail, thyroid adenomas (TAs) are benign tumors, most commonly of the follicular type, characterized by regulated growth and a lack of invasive potential. Metabolically, adenomas exhibit a moderate glycolytic activity that is lower than that observed in carcinomas, with partial preservation of oxidative phosphorylation (OXPHOS) indicating a less pronounced Warburg effect (33,34). These lesions typically retain the expression of the sodium/iodide symporter (NIS), allowing iodine uptake that remains useful for diagnostic imaging, such as I¹³¹ scintigraphy. Mitochondrial function is generally preserved, ensuring efficient ATP production. At the genetic level, adenomas often harbor mutations in genes such as RAS or PAX8/PPARγ, which promote cellular proliferation without inducing malignant transformation. These mutations can activate signaling pathways like MAPK and PI3K/Akt, though without triggering the extensive metabolic reprogramming seen in thyroid cancers (35). Additionally, slight upregulation of enzymes involved in lipid and amino acid metabolism, such as fatty acid synthase and the glutamine transporter ASCT2, has been reported, though to a lesser extent than in malignant tumors (36,37).

In contrast, TCs, which include differentiated forms (papillary, follicular), as well as medullary and anaplastic subtypes, exhibit a reprogrammed metabolism that supports rapid growth, invasion, and metastatic spread. These malignant tumors demonstrate increased glucose consumption, driven by the overexpression of glucose transporters GLUT1/3 and glycolytic enzymes such as hexokinase-II and lactate dehydrogenase A, leading to high lactate production even under aerobic conditions, a hallmark of the Warburg effect. This metabolic phenotype is particularly evident in papillary carcinomas with the BRAF V600E mutation and in anaplastic carcinomas. The high glycolytic rate is also reflected in positron emission tomography (PET) imaging, where elevated 18F-fluorodeoxyglucose (¹⁸F-FDG) uptake is commonly associated with tumor aggressiveness (38,39).

Mitochondrial dysfunction is frequently observed in advanced carcinomas, especially in anaplastic forms, where there is downregulation of mitochondrial genes such as COX7A1 and loss of NIS function, resulting in resistance to radioiodine therapy (40). Nevertheless, certain follicular carcinomas may retain OXPHOS capacity, particularly in the presence of RAS mutations, utilizing alternative substrates such as glutamine and fatty acids.

Iodine metabolism is also markedly altered in carcinomas. The expression of NIS is often lost due to epigenetic or mutational mechanisms, impairing the efficacy of I¹³¹-based radiometabolic therapy. While some differentiated carcinomas, such as papillary types, may retain NIS expression, its distribution is often heterogeneous.

The tumor microenvironment in advanced carcinomas is characterized by hypoxia-induced activation of hypoxia-inducible factor 1-alpha (HIF-1α), which promotes pro-angiogenic signaling via VEGF and further enhances glycolytic metabolism, contributing to increased invasiveness (40). Moreover, metabolic symbiosis may occur between cancer cells and surrounding stromal elements, such as astrocytes, which can provide lactate as an alternative energy source through a reverse Cori cycle.

Genetically, distinct metabolic alterations are associated with specific carcinoma subtypes. Papillary carcinomas harboring BRAF V600E mutations show enhanced MAPK pathway activation along with increased HIF-1α expression and glycolysis (35).

Despite these promising results, our study has some limitations. Firstly, the study sample is small, particularly regarding the carcinoma and adenoma cases. This limitation precluded the application of AI models, which require large datasets for effective training, validation, and test. It is well known, in fact, that AI applied to thermal images of the neck can help to detect thyroid cancer (26,30,31,41). For instance, Chantasartrassamee et al., reached an accuracy rate of 98.8% and a sensitivity of 96.3% when identifying thyroid malignancies (25), demonstrating the effectiveness of combining AI with IRT images to discriminate tumor cases. Indeed, future research should expand the study cohort and integrating AI-based pattern recognition approaches to develop predictive algorithms capable of classifying thyroid nodules with high accuracy. Additionally, future studies could explore the use of dynamic thermal responses (e.g., after a controlled cooling stimulus) to enhance the sensitivity of IRT. Notably, a validated model linking the skin temperature of the neck with the tumor typology would enable physician to monitor thermal changes in thyroid nodules longitudinally during active surveillance protocols, reducing reliance on repeated imaging or biopsies.

A further limitation concerns the ROI definition, which was based on anatomical landmarks at the level of the whole lobe rather than on a segmentation of the individual nodule. While this choice makes the procedure fast and reproducible, and is consistent with a screening-oriented application, it may dilute focal thermal alterations produced by small lesions within the averaged signal. Future studies integrating ultrasound co-registration or automatic segmentation of the glandular contour could refine the spatial specificity of the measurement.

Finally, it is worth noting that bilateral cases can affect the effectiveness of the method. However, the effect of bilateral disease is confined to sensitivity. In fact, since all such patients belong to the positive class, specificity is unaffected. Moreover, the magnitude of the dilution is a function of the symmetry of the thermal burden rather than of the mere presence of bilateral lesions, thus, complete cancellation of the signal requires two lesions of comparable size, vascularization and metabolic activity, symmetrically positioned, which is uncommon, so that a partial asymmetry usually persists and the dilution is generally incomplete. However, metrics that do not rely on lateral symmetry (such as the absolute temperature of each lobe normalized to a reference region of the same subject, the intra-lobar thermal heterogeneity of a single ROI (standard deviation, spatial gradient, hot-spot area), or comparison against a normative model derived from a healthy population) would remain informative when both lobes are involved and could be used in combination with the differential metric. Hence, further studies should be performed to develop and validate such complementary indices, in a cohort large enough to include an adequate number of bilateral cases.

Furthermore, validating IRT performance in outpatient or non-surgical settings would be crucial to assess its reliability under less controlled environmental conditions, thereby supporting its potential for routine clinical use.

Conclusions

This pilot study reports that IRT represents a novel, safe, effective, and low-cost tool that could be validly employed to assist the physician in the diagnostic process of thyroid tumors.

Due to the sample size limitations of our study, the current data we present have an explorative scope.

The primary finding of the paper relies on the possible capability of IRT to identify the uncertain cases represented by the TIR 3A and TIR 3B of the TIR classification system of the FNAC results, providing possible important indications for the surgical treatment of indeterminate nodules. Moreover, the correlation between the external temperature and the inner temperature (i.e., skin of the neck and exposed thyroid) was investigated, disclosing a different behavior for the adenomas and carcinomas, supporting the hypothesis that superficial thermal images can be indicative of the metabolic and vascular condition of the thyroid associated with the pathology.

However, despite the limited sample size, particularly in the TIR 3A and TIR 3B subgroups, the proposed methodology is well-defined and reproducible. Further investigations involving larger cohorts are indeed necessary to better define IRT’s real potential to discriminate among different thyroid tumor types, also in less controlled environmental conditions, and to support the development of a reliable and easily applicable IRT-based model for clinical practice.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of the “G. d’Annunzio” University of Chieti-Pescara, with approval covering the research activities conducted at the Casa di Cura Pierangeli Hospital in Pescara (approval No. Prot. 392490/24, 09/10/2024 - C.Et.R.A. Pescara). All patients gave their informed consent prior to their inclusion in the study.

Data Sharing Statement

Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1482/dss

qims-16-11-875-dss.pdf ^{(116.3KB, pdf)}

DOI: 10.21037/qims-2026-1482