Breast cancer is the most common malignancy among women worldwide, with one in eight women
receiving this diagnosis during their lifetime. Surgery remains the cornerstone of treatment, yet
despite its curative potential, it often carries profound psychological consequences due to feelings of
loss and deformity.
Breast reconstruction plays a pivotal role in restoring form and well-being. Options are broadly
classified into alloplastic (implant-based) and autologous (tissue-based) techniques. The current gold
standard for autologous reconstruction is the DIEP (Deep Inferior Epigastric Perforator) flap, which
uses abdominal tissue and requires microsurgical anastomosis between the deep inferior epigastric
vessels and the internal mammary vessels.
Successful flap harvest depends on accurate identification and preservation of the dominant
perforator supplying the flap. Pre-operative CT angiography (CTA) is routinely used to map abdominal
perforators, but interpretation is time-consuming, requires specialist expertise, and is prone to human
error.

Project Aim
This project seeks to harness artificial intelligence (AI) to enhance perforator detection in preoperative
planning for microsurgical breast reconstruction. By automating vessel identification from
CTA imaging, AI could reduce planning time, improve precision, and standardize decision-making,
ultimately optimizing patient outcomes.

The primary objective is to evaluate the role of AI in automated perforator selection for free flap
surgery.
Specific goals include:
- Developing an AI-driven system to detect and characterize perforators from radiographic
imaging.
- Integrating these findings into the surgical workflow to assist surgeons and radiologists.
- Assessing efficiency gains, accuracy, and clinical relevance compared with current practice.
Hypothesis: Incorporating AI into perforator detection will significantly improve efficiency and
accuracy of pre-operative planning in microsurgical breast reconstruction.

This is an ongoing project where we have already had 1 Master student in 2024-2025 (Automated
segmentation), 1 currently in 2025-2026 (automating perforator coordinates and AI development)
- The first phase established segmentation and computer-aided detection of perforators.
- This summer, an umbilicus detector was developed, replacing landmark regression with a UNet
+ heatmap regression approach.
- This year, the validating the CAD system on patients to assess accurate quantitative perforator
detection, automating perforator coordinates which are compared to the coordinates
identified by the radiologist (clinical comparison of automated CAD)
- The next phase will focus on refining predictive modelling to not only map vessels
quantitatively but also rank the most favorable perforators for clinical application.
Achievements to Date (2024–2026)
- Comprehensive literature review on AI applications in medical imaging and perforator flap
surgery.
- State-of-the-art analysis of vascular detection techniques.
- Development of a CAD pipeline for perforator identification.
- Publication and presentation of 1 conference article (MICCAI Deep Breath) and 1 journal
article that is In Press (Artificial Intelligence Surgery)
Objectives for 2026–2027
The upcoming phase will emphasize refinement and clinical translation.
Planned milestones include:
- Improving AI accuracy in mapping the perforator’s course and arborization within
subcutaneous tissue.
- Developing a decision-support tool incorporating key perforator characteristics i.e. AI
providing the best perforators based on:
- Vessel size
- Precise anatomical location
- Arborization pattern within subcutaneous fat
- Intramuscular course and depth
By integrating these features, the project aims to streamline pre-operative planning, reduce
intraoperative uncertainty, and enhance surgical outcomes in microsurgical breast reconstruction.