Automated vascular analysis of breast thermograms with interpretable features

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Vascular changes are early indicators of breast cancer.
• Thermal imaging offers a low-cost method for visualizing vascularity.
• Manual interpretation of thermal vascularity is subjective and challenging.

Purpose of the Study:

• To develop automated techniques for breast thermal image analysis.
• To assist physicians in visualizing and interpreting vascularity for malignancy classification.
• To quantify vascular features for improved diagnostic accuracy.

Main Methods:

• A deep learning-based semantic segmentation approach was used to generate vessel heatmaps.
• Interpretable vascular parameters were extracted from thermal images.
• A classifier was developed to predict breast cancer likelihood using these parameters.

Main Results:

• The approach successfully segmented breast vessels.
• A classification performance with an area under the receiver operating characteristic curve of 0.85 was achieved.
• Validation was performed on an independent clinical dataset of 258 participants.

Conclusions:

• The developed method effectively detects vasculature in breast thermal images.
• The approach demonstrates utility in the interpretation of breast vascularity for cancer detection.
• High classification performance indicates potential for clinical application.