Clinical Feasibility of Low-Magnification Fluorescence Imaging for Breast Cancer Margin Detection Using Texture Analysis and Deep Learning

This study investigated how 4× and 10× magnifications affect margin-level classification of MUSE images using both TA and DL methods, employing a patch-level framework. For TA, the red channel of each patch was preprocessed, LBP features were extracted, and an SVM predicted the patch-level class. For DL, patch embeddings were processed through a ViT with an MLP head to get patch-level predictions. In both methods, patch predictions were combined and compared to a binary threshold to predict the final MUSE image prediction. Quantitative results showed that DL and TA performed consistently across both magnifications, though TA was slightly less accurate. Overall, 10× magnification did not improve margin-level accuracy compared to 4×. Since 10× takes longer to acquire and requires more computation, 4× is a practical and efficient choice for intraoperative margin assessment. Overall, MUSE has the potential to require less time than traditional histopathology methods, such as frozen-section analysis, and to provide diagnostic results using DL, thereby reducing the need for manual image interpretation by pathologists. These findings can help develop more accurate diagnostic models, support careful margin inspection during breast-conserving surgery, and improve patient outcomes.

Improvements to AI Systems:

1. Magnification-Agnostic Training: Train the AI to be invariant to magnification by using multi-scale data augmentation (e.g., randomly resizing patches between 4× and 10× during training). This reduces the need for separate models per magnification and improves robustness across different clinical scanners.

2. Computational Efficiency Optimization: Implement a dynamic resolution selection mechanism where the AI first analyzes low-resolution (4×) patches and only upscales to 10× for ambiguous or high-uncertainty regions (e.g., near the margin boundary). This cuts acquisition and inference time while maintaining accuracy.

3. Uncertainty-Aware Margin Classification: Add a confidence score to each patch-level prediction (e.g., using Monte Carlo dropout or ensemble variance). The AI can then flag low-confidence patches for pathologist review, reducing false negatives in critical margin areas.

4. Cross-Method Ensemble Learning: Combine traditional analysis (LBP + SVM) with deep learning (ViT) at the patch level using a learned fusion layer. This leverages the interpretability of TA and the feature richness of DL, improving overall accuracy without extra magnification cost.

5. Real-Time Intraoperative Decision Support: Deploy the improved AI as a real-time tool that processes MUSE images at 4× magnification, overlays margin-risk heatmaps, and provides a binary clear vs. involved margin call within seconds. This reduces reliance on frozen-section histopathology and shortens surgery time.

What the Improved AI System Can Do:

• Accurately classify breast tissue margins from MUSE images at 4× magnification with performance equal to 10×, but with 2–3× faster acquisition and lower computational load.

• Automatically adapt to different magnification inputs without retraining, making it deployable across varied clinical hardware.

• Provide per-patch confidence scores, enabling pathologists to focus only on uncertain regions, thereby reducing manual review workload.

• Integrate both handcrafted and learned features to achieve higher sensitivity for detecting positive margins, potentially lowering re-excision rates.

• Operate in real-time during breast-conserving surgery, giving surgeons immediate feedback on margin status and improving patient outcomes by reducing the need for second operations.