Development and educational evaluation of patient-specific 3D-printed breast phantoms for oncoplastic surgical training

Abstract

Background

Patient-specific 3D-printed breast models provide a high-fidelity, ethically sound alternative to traditional training platforms for oncoplastic surgery. They can improve anatomical understanding and procedural proficiency while reducing reliance on cadaveric specimens. Their reusability and customization potential support cost-effective, sustainable programmes, particularly in resource-limited environments.

Methods

Two patient-specific breast models with embedded tumor lesions were fabricated using multimaterial PolyJet 3D printing and silicone post-processing, based on MRI datasets from two patients, one imaged prone and one supine. Six breast and reconstructive surgeons tested the models in a simulated surgical environment, performing training scenarios including mastectomy flaps, tumor excision, and perforator identification, and completed a structured 7-item questionnaire (5-point Likert scale) assessing usability, anatomical realism, educational value, and overall satisfaction; an overall usability metric was derived through a System Usability Scale (SUS)-like calculation.

Results

Surgeons reported high satisfaction regarding haptic realism and ease of use, with most items rated 4–5 on the Likert scale. The models enhanced procedural confidence, spatial understanding of breast anatomy, and educational engagement, and yielded a SUS-like score of 71.43, indicative of a positive user experience. Reusability and portability facilitated repeated practice without biosafety concerns, while durability showed greater variability across raters.

Conclusions

Patient-specific 3D-printed breast phantoms provide a feasible, ethical, and educationally valuable platform for oncoplastic surgical training. This study suggests their potential to support skill acquisition and standardized training, and to reduce reliance on cadaveric specimens, supporting sustainable, cost-effective and pathology-integrated educational programs. However, larger, outcome-based studies are needed to confirm educational efficacy.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12893-026-04110-9.

Background

Breast surgeries, including both oncologic and reconstructive procedures, necessitate a high degree of precision and an in-depth understanding of anatomical variability. Conventional training paradigms predominantly rely on supervised clinical practice, cadaveric dissections, and artificial models, each presenting inherent limitations. Training on live patients poses significant ethical concerns, cadaveric specimens lack dynamic biomechanical properties of living tissue and generally do not contain tumors or other pathologies, and synthetic models often fail to replicate anatomical accuracy and tissue heterogeneity required for realistic simulations.

Silicone-based composite materials have been employed to replicate breast tissue properties for ultrasonography training, providing improved tactile and imaging realism [1]. Additionally, specialized simulators for sentinel lymph node biopsy have established new benchmarks in surgical education by enhancing procedural accuracy and skill acquisition [2].

Nevertheless, despite these advances, the overall fidelity and pathology-specific realism of traditional simulators remain limited. The acquisition of surgical proficiency by young surgeons is a complicated challenge, demanding a delicate balance between educational imperatives and patient safety. The traditional apprenticeship model, while historically central to surgical training, presents limitations in contemporary practice. The inherent variability in case availability and the progressive nature of skill transfer can result in inconsistent learning experiences and prolonged training durations [3, 4]. Furthermore, the ethical considerations surrounding patient safety during the initial stages of surgical training necessitate a cautious approach, potentially hindering the rapid development of independent surgical competence.

On the other hand, cadaveric dissection offers an anatomically precise training modality, enabling trainees to familiarize themselves with complex anatomical structures and surgical procedures. However, the use of cadaveric specimens is not without its challenges. Fixation and preservation techniques, while necessary, can alter tissue biomechanics, thereby diminishing the fidelity of surgical simulation compared to living tissue. Moreover, logistical constraints, including accessibility, storage, and the inherent costs associated with cadaveric procurement and maintenance, limit the widespread adoption of this training method [3].

Cadaveric specimens also lack pathological features such as tumors, further reducing their relevance for oncologic or pathology-specific surgical training. Additionally, the ethical considerations surrounding the use of cadaveric material are a significant factor.

Commercially available synthetic surgical simulators provide a standardized and repeatable training environment. However, these models often fail to accurately replicate the nuanced biomechanical properties of human tissues, the intricate vascular architecture, the inherent anatomical variations observed in clinical practice, and they do not incorporate patient-specific pathologies such as tumors. This lack of realism can impede the development of critical tactile skills and spatial reasoning essential for surgical mastery [5, 6].

The advent of three-dimensional (3D) printing technology offers a promising avenue to address these limitations. By leveraging high-resolution imaging data, such as magnetic resonance imaging (MRI), 3D printing enables the fabrication of patient-specific anatomical models with enhanced fidelity. The utilization of advanced materials, including silicone composites and polymer-based resins, facilitates the replication of soft tissue mechanics [7, 8], thereby improving the realism of surgical simulation. Furthermore, 3D printing allows for the incorporation of intricate vascular structures and anatomical variations, as well as tumors or other pathologies, providing a more comprehensive training experience. The diverse applications of 3D printing in surgical training, including the creation of simulation models for complex procedures such as oncoplastic breast surgery and reconstructive techniques, highlight the potential of this technology to bridge the gap between traditional training methods and the demands of contemporary surgical practice [7–15].

Notably, most existing breast training models share two recurring limitations: they rarely incorporate patient-specific pathology, such as an embedded tumor lesion, and they are typically derived from a single imaging position, overlooking the substantial deformation the breast undergoes between the prone (imaging) and supine (operative) setups. The present workflow addresses both aspects, integrating a patient-specific tumor lesion and supporting the fabrication of either a prone- or a supine-derived model from the same pipeline, so that the configuration most relevant to a given training objective can be selected.

Building upon these premises, and to address the two limitations outlined above, the aim of this study was to design and validate two patient-specific 3D-printed breast phantoms incorporating tumor structures, one derived from prone and one from supine MRI, and to assess their educational impact on oncoplastic surgical training through structured expert user feedback.

Methods

Study design and imaging data

This study followed a multi-step workflow designed to create anatomically faithful, patient-specific 3D breast models from medical imaging data, enabling realistic simulation for surgical training (Fig. 1). The process comprised: (1) acquisition of high-resolution MRI datasets in either the prone or the supine positions of pathological anatomies; (2) segmentation and 3D reconstruction of anatomical regions of interest; (3) post-processing for mesh optimization and multimaterial printing preparation; (4) PolyJet-based additive manufacturing; and (5) manual post-processing with silicone casting to reproduce external anatomical features. Each phase was optimized to ensure high anatomical fidelity, mechanical realism, and visual clarity. Following fabrication, the 3D models were used by clinicians during educational activities and events to simulate surgical techniques, after which surgeons completed a structured feedback questionnaire to assess usability, educational value, and perceived impact on surgical planning.

Fig. 1.

Overview of the multi-step workflow for producing the patient-specific breast phantoms: (1) acquisition of prone or supine MRI; (2) segmentation and 3D reconstruction; (3) mesh optimization and multimaterial print preparation; (4) PolyJet additive manufacturing; (5) manual post-processing with silicone casting

Imaging data were obtained from two female patients (patient 1: AC; patient 2: PC), each patient was imaged at a single institution and in a single position. Patient 1 (AC) was a 47-year-old female affected by a T1 N0 luminal A breast cancer located in the upper outer quadrant of the left breast, who underwent a prone MRI scan at Humanitas Istituto Clinico Catanese (HICC), Catania, Italy. Her BMI was 22.5; nipple-to-sternal-notch distance was 22 cm; ptosis according to Regnault was G0. Patient 2 (PC) was a 62-year-old female with a right cT1 N0 luminal B breast cancer located in the lower outer quadrant; nipple-to-sternal-notch distance was 25 cm, ptosis G3 and BMI 29, admitted at Champalimaud Foundation (CF), Lisbon, Portugal, where she underwent a supine MRI.

All MRI scans were anonymized prior to processing to protect patient privacy. Prone MRI data were acquired at HICC (Philips Achieva dStream 1.5 T), while supine MRI data were provided by CF (Philips MR7700 3 T). Consent was obtained according to the regulations of the local institutions participating in the study: at HICC, the patient consented to the use of anonymized images for research purposes with a specific form at the time of admission; at CF, the image transfer was approved by the internal ethics committee.

Each phantom was therefore reconstructed and printed in the same position in which its source MRI had been acquired, yielding two distinct models: a prone-derived phantom for Case 1 and a supine-derived phantom for Case 2. In this way, a single workflow can generate either configuration from the corresponding acquisition, so that the position most relevant to a given training objective can be selected.

Imaging acquisition and segmentation

As indicated above, each anatomical dataset was acquired using MRI scans either in prone (Philips Achieva dStream 1.5 T) or supine (Philips MR7700 3 T) positions. The prone imaging captured anatomical displacement under gravity for surgical planning, while the supine position simulated the typical operative setup. Segmentation was performed using Materialise Mimics v27.0 (Materialise, Leuven, Belgium), a medical image processing software. Each MRI dataset was explored in axial, sagittal, and coronal views to accurately identify and delineate key anatomical structures (Figs. 2 and 3).

Fig. 2.

Segmentation of Case 1 (AC) from prone MRI

Fig. 3.

Segmentation of Case 2 (PC) from supine MRI

Segmented structures included: rib cage, glandular and adipose tissue, major vascular components, and the tumor lesion. Color-coded overlays were assigned to each structure for clarity during processing (see Fig. 4a and b).

Fig. 4.

3D reconstruction of breast anatomy for the two cases: a Case 1 — rib cage (violet), glandular tissue (light blue), blood vessels (green), tumor (yellow); b Case 2 — rib cage (pink), sternum (yellow), blood vessels (red), glandular tissue (light blue), tumor (orange)

3D modelling and printing

Segmented volumes were exported as STL files and refined using Materialise 3-matic v19.0 (Materialise, Leuven, Belgium) for mesh correction, smoothing, and digital reconstruction of anatomical continuity.

Models were printed on a Stratasys J850 Digital Anatomy Printer (Stratasys Ltd., Rehovot, Israel) using PolyJet technology, with print jobs prepared in GrabCAD Print. This platform was selected because it supports the Digital Anatomy material family (TissueMatrix, GelMatrix) and the associated library of validated anatomical presets, and because it allows the simultaneous deposition of multiple materials with different mechanical properties and colours within a single build.

Material assignment (Table 1) was based on the manufacturer’s validated anatomical presets for each tissue class, refined according to the previous experience of our group in the fabrication and clinical evaluation of 3D-printed anatomical models, in which these combinations had been qualitatively assessed by clinicians and had consistently provided satisfactory tactile behaviour. The governing criteria were the preservation of a perceivable stiffness gradient between glandular and adipose tissue, an adequate contrast between soft tissue and the tumor lesion, and printability constraints, in particular the minimum wall thickness required to reproduce vessels down to 3 mm in diameter.

• Glandular tissue was reproduced with a 2 mm outer layer of Agilus30 White over a 60/40 mix of TissueMatrix and GelMatrix.

• Adipose tissue used a thinner outer layer and a reversed internal ratio for lower stiffness.

• The rib cage and tumor were printed using predefined “very stiff” and “non-encapsulated tumor” presets.

• Blood vessels were printed using Agilus Magenta to simulate vascular pliability and visibility down to diameters of 3 mm.

Table 1.

Summary of material formulations for each anatomical region. The selected material combinations were optimized to balance haptic fidelity, mechanical realism, and printability

Anatomical region	Material type/Outer layer	Background
Glandular Tissue	2 mm of Agilus30 White	40% GelMatrix, 60% TissueMatrix
Adipose Tissue	0.3 mm of 50% Agilus White and 50% TissueMatrix	60% GelMatrix, 40% TissueMatrix
Rib Cage	Musculoskeletal Ribs Very Stiff preset	–
Tumor	General Anatomy – Solid Tumor Stiff Non-Encapsulated Tumor preset	General Anatomy – Solid Tumor Stiff Non-Encapsulated Tumor preset
Blood Vessels	Agilus Magenta	Agilus Magenta

Post-processing and assembly

Following printing, support material was removed manually and with the aid of a waterjet before any silicone work. Support material inside the vascular lumina, printed with GelMatrix, was cleared by water flushing combined with manual removal. The models were then rinsed thoroughly and left to air-dry at room temperature until completely dry, to ensure that no support residue remained on the surfaces to be bonded prior to silicone casting (Fig. 5).

Fig. 5.

Final anatomical print of the breast (from prone imaging, Case 1)

Pectoralis major and intercostal muscles were reconstructed using Ecoflex-10 silicone for enhanced haptic realism. Skin and nipple structures were cast using customized moulds and the same silicone material (Fig. 6). The moulds used for casting the skin and nipple structures were obtained as the negative of the segmented external surface and fabricated by fused deposition modelling in PLA on an Original Prusa i3 MK3S (Prusa Research, Prague, Czech Republic). The resulting model preserved distinct visual and mechanical zones, allowing for tactile discrimination between anatomical compartments.

Fig. 6.

Muscles and skin fabricated using silicone for (a) supine and (b) prone models

Educational simulation and evaluation

The patient-specific 3D-printed breast models were evaluated during educational sessions, where the models were placed on tables configured to simulate the surgical environment (see Figs. 7 and 8). The evaluation involved six surgeons (four breast consultants and two senior fellows) with a median of 12 years (range 5–24) of breast surgery experience. Four routinely performed level II oncoplastic procedures (>50/year), whereas two had intermediate oncoplastic experience. Five participants had previous exposure to surgical simulation models, although none had previously used the presented phantom. All participants completed the same standardized sequence of tasks: mastectomy flap dissection, tumour excision, and perforator identification, using identical instruments after a standardized briefing. Each evaluation session lasted approximately 30 minutes (5-minute introduction, 20-minute hands-on evaluation, 5-minute questionnaire).

Fig. 7.

Hands-on training performed on the Case 1 printed phantom

Fig. 8.

Image from live surgical simulation performed on the Case 2 printed phantom

After the sessions, a structured 7-item questionnaire assessing usability, anatomical realism, educational value, and overall satisfaction on a 5-point Likert scale was administered to all six surgeons. The seven-item questionnaire was developed in-house for the present study; the full English version is provided in the Supplementary Files. To provide an overall usability metric, a System Usability Scale (SUS)-like score was calculated: although the standard SUS comprises 10 items [16], the calculation was adapted to the 7-item questionnaire used in this study.

Collected data were subsequently aggregated and analyzed to assess overall user satisfaction and to identify potential areas for model improvement. Reference values employed for the interpretation of SUS scores are those reported by Bangor et al. [17].

Results

The resulting 3D-printed phantoms demonstrated high anatomical accuracy and mechanical realism, as validated by expert observers including breast surgeons. Haptic feedback was consistently rated as comparable to cadaveric tissue, especially for glandular versus adipose differentiation. Visual clarity, enabled by multicolor, multimaterial PolyJet printing, allowed immediate identification of anatomical structures, and vascular anatomy was accurately reproduced down to 3 mm in diameter, supporting practice of perforator dissection.

Functional realism was confirmed by the use of the phantoms in complete training scenarios, including mastectomy flaps, perforator identification and tumor excision with appropriate tactile cues, allowing microsurgical instrument handling without compromising structural integrity (Fig. 7). Surgeons noted improved spatial orientation and confidence in flap planning due to clear visualization of internal anatomical relationships. A qualitative survey indicated high satisfaction scores (>90%) regarding educational value, and simulation engagement (Fig. 8).

Overall, the models received high ratings across all assessed domains. Most participants rated the models as highly useful for educational and training activities, effective for educational activity, and useful for discussion with colleagues (predominantly scores of 4–5). Ease of use was generally rated positively, while durability showed more variability, with some participants indicating lower scores (1–3). Notably, participants reported that the models contributed to improved learning outcomes, with the majority giving scores of 4–5 (Fig. 9).

Fig. 9.

Diverging stacked bar chart of clinician feedback, showing the distribution of Likert-scale responses (1–5) across the evaluated categories (overall usefulness for education, usefulness for preparing/delivering educational activity, usefulness for discussion, effectiveness for educational activity, ease of use, durability, and improvement in learning outcomes)

The diverging stacked bar chart illustrates the distribution of responses for each questionnaire item, highlighting the proportion of negative, neutral, and positive feedback.

We computed an exploratory, SUS-inspired usability index adapted to the seven items; the resulting value (71.43) should be interpreted with caution given the small sample and the deviation from the validated 10-item instrument, but is consistent with the positive qualitative feedback. These findings suggest good usability and perceived educational value of the developed phantoms, as reported by expert users.

Discussion

The feedback received from the six surgeons clearly highlights the potential of the models as efficient practical tools for surgical education. The SUS-inspired usability index complements the qualitative feedback and aligns with the perceived usefulness reported by the surgeons, in line with previous studies on the value of patient-specific 3D-printed models in training.

The standardized fabrication of the models enables consistent educational comparisons, allowing different surgeons to be tested on the same anatomical phantom, or alternatively the same surgeon to train across different patient-specific models. The reusability of the phantoms and the lack of biosafety requirements facilitated storage at room temperature and easy transport in standard cases, making them suitable for off-site or modular courses. Moreover, this flexibility could enhance the reproducibility of skill assessment and support a more standardized evaluation of surgical performance in future studies.

For instance, Ustbas et al. [1] demonstrated that silicone-based composite models can replicate the biomechanical characteristics of mammary tissue, providing a reproducible and cost-effective platform for ultrasound-guided training. Likewise, Keshtgar et al. [2] developed a sentinel lymph node biopsy simulator that reproduced key surgical steps and improved procedural precision. In the context of oncoplastic training, enhanced trainee competency in mammoplasty and flap-based reconstruction when using silicone-based models was reported by Kilic et al. [5], and further studies confirmed that patient-specific phantoms can improve surgical efficiency and reduce intraoperative errors [4, 18]. Extending beyond procedural training, it has been shown that 3D-printed or imaging-derived models optimize preoperative planning and reduce operative time, particularly in perforator flap reconstruction [19, 20]. Lastly, Mayer et al. [14] emphasized the potential of 3D printing for patient-specific implants and biodegradable scaffolds, marking a step toward personalized surgical solutions.

Compared with previously reported breast training models, the present approach offers two distinctive features: the integration of a patient-specific tumor lesion and the ability to derive, from a single workflow, either a prone- or a supine-based model, which support pathology-oriented and position-aware rehearsal. Conversely, it currently lags behind some established platforms in other respects, most notably the absence of ultrasound compatibility. Positioning the model along these axes clarifies both its added value and the directions in which it still needs to mature.

The choice of fabrication technology reflects a deliberate, value-driven trade-off rather than a single-technology approach. Alternative additive technologies would have reduced cost substantially: fused deposition modelling (FDM) and stereolithography (SLA) are widely used for patient-specific anatomical models and achieve comparable dimensional accuracy [21], at a fraction of the material cost [22]. However, both are essentially single-material, rigid-output processes: they cannot, within a single build, reproduce the graded stiffness and multi-tissue differentiation (glandular, adipose, vascular, tumoral) that support the haptic realism pursued here. Conversely, silicone-only casting can deliver realistic soft-tissue behaviour but does not natively provide patient-specific internal architecture without an additive step to generate the moulds or cores.

Rather than committing the entire model to a single process, we deliberately matched each component to the most appropriate technology. PolyJet Digital Anatomy printing was reserved for the internally complex, multi-tissue core, where the simultaneous reproduction of glandular, adipose, vascular and tumoral structures with distinct mechanical properties genuinely requires multi-material additive manufacturing and justifies its higher cost. For the external soft-tissue envelope, where this level of internal differentiation is not needed, we instead used silicone casting supported by low-cost FDM-printed moulds, achieving realistic surface behaviour at a small fraction of the cost. The whole phantom could, in principle, have been produced on the Digital Anatomy platform, but doing so would have increased cost substantially without a corresponding gain in realism for these outer regions. This hybrid strategy therefore concentrates the expensive technology where it adds unique value and relies on cheaper, equally adequate solutions for the remainder.

Cost nonetheless remains the principal barrier to wider adoption: among additive options, PolyJet Digital Anatomy printing lies at the higher end of the spectrum, driven mainly by the price of the multi-material photopolymers and by the support material consumed. A detailed cost analysis was beyond the scope of this study and represents a relevant direction for future work aimed at industrialisation.

A further consideration concerns the nature of the evaluation itself. The structured questionnaire was the only quantitative measure collected in this study: procedural times were not recorded during the sessions, and no validated performance instrument was administered. The assessment presented here therefore addresses the face and content validity of the phantoms - whether expert users judge them anatomically and tactilely credible and educationally useful - rather than the acquisition or transfer of surgical skill.

This is consistent with the composition of the panel: the participants were experienced breast surgeons and senior fellows, who are appropriate judges of anatomical realism and educational potential but are not the population in which skill acquisition can be meaningfully measured. Establishing educational efficacy would require a different design, based on trainees rather than experts and incorporating objective measures. A pre-post design on a cohort of residents and fellows, with blinded rating of recorded performances, would allow the educational value of these phantoms to be tested, and represents the natural next step of this work.

Further limitations concern the models themselves and the scale of the study, which remains preliminary and exploratory, with a small number of expert evaluators. Refinements in current 3D-printed materials are needed to more accurately simulate the nuanced textural properties of skin and the viscoelastic behavior of glandular tissue, consistent with the variability observed in the durability ratings. The current models also omit anatomical structures relevant to other aspects of reconstructive planning, such as nerves and the ligamentous structures; these were left out because the present phantoms were designed around the specific training scenarios targeted in this study (mastectomy flaps, perforator identification and tumor excision) which rely on vascular, glandular, adipose and tumoral anatomy. Their inclusion is a worthwhile direction for extending the anatomical completeness of the phantoms to a broader range of reconstructive tasks. The effect of colour coding was not evaluated, and reproducing more realistic tissue colours is a straightforward future refinement.

Finally, improvements in ultrasound compatibility are essential to more closely replicate clinical scenarios, and financial barriers associated with the adoption of these technologies restrict their widespread implementation in educational programs.

As highlighted by Shichinohe et al. [3], the ethical and logistical limitations inherent in cadaver-based training underscore the necessity of developing high-fidelity, artificial models as a sustainable and scalable alternative in contemporary surgical pedagogy.

Conclusions

Patient-specific 3D-printed breast phantoms provide a feasible, ethical, and educationally valuable platform for oncoplastic surgical training. They may facilitate skill acquisition and offer a standardizable, reproducible alternative that reduces reliance on cadaveric specimens, supporting safe rehearsal of high-stakes procedures such as perforator flap harvesting and mastectomies. By bridging the gap between conventional simulators and real operative experience, these models have the potential to improve surgical education, preoperative planning, and ultimately patient outcomes, representing a promising addition to the available tools for oncoplastic and reconstructive education. Beyond the breast, the same imaging-to-phantom workflow is, in principle, transferable to other anatomical districts and pathologies, suggesting a broader role for patient-specific, pathology-integrated models in surgical education. Future research should focus on augmented reality integration, improved ultrasound compatibility, outcome-based validation on larger cohorts of trainees, and standardized production protocols to ensure broad adoption and long-term impact.

Abbreviations

3D

Three-dimensional

MRI

Magnetic resonance imaging

STL

Standard Tessellation Language (stereolithography file format)

BMI

Body mass index

SUS

System Usability Scale

HICC

Humanitas Istituto Clinico Catanese

CF

Champalimaud Foundation

Authors' contributions

GC and PO conceived the study and designed the workflow, CC and AB wrote the first draft of the manuscript. PG and AC contributed to data acquisition and to the educational activities. NKM and CC developed the phantoms. SM contributed to the production of the models. All authors contributed to manuscript revision and read and approved the final manuscript.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability

The datasets generated and analyzed during the current study are not publicly available due to patient confidentiality, but anonymized datasets and 3D model STL files are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the 1964 Helsinki Declaration and its later amendments. Imaging data were obtained from two patients: at Humanitas Istituto Clinico Catanese (HICC), the patient provided written consent for the use of anonymized images for research purposes at the time of admission; at the Champalimaud Foundation (CF), transfer of the anonymized images was approved by the institution’s internal ethics committee.

Consent for publication

Consent for publication was obtained from all participants.

Competing interests

The authors declare that they have no competing interests.