Related Experiment Video
Updated: Sep 17, 2025
10:37
A Multimodal Imaging Framework to Advance Phenotyping of Living Label-free Breast Cancer Cells
Published on: August 22, 2025
283
An explainable-by-design end-to-end AI framework based on prototypical part learning for lesion detection and
Andrea Berti1,2,3, Camilla Scapicchio2,4, Chiara Iacconi5
1Institute of Information Science and Technologies (ISTI) - National Research Council of Italy (CNR), Via Giuseppe Moruzzi, 1, Pisa, 56127, Italy.
Computational and Structural Biotechnology Journal
|July 2, 2025
Summary
This study introduces a transparent AI system for analyzing Digital Breast Tomosynthesis (DBT) scans, improving early breast cancer detection. The AI provides accurate predictions with explanations, aiding radiologists and enhancing patient outcomes.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Healthcare
- Oncology
Background:
- Breast cancer is a leading global cancer in women, necessitating effective early detection methods.
- Digital Breast Tomosynthesis (DBT) offers improved clarity over mammography but generates large datasets challenging timely analysis.
- Current AI in medical diagnostics often lacks transparency, hindering clinical trust and adoption.
Purpose of the Study:
- To develop and evaluate a transparent Artificial Intelligence (AI) system for the expedited and interpretable analysis of DBT scans.
- To enhance the accuracy and efficiency of breast cancer lesion detection and classification using AI.
- To provide explainable predictions that support clinical decision-making in breast cancer screening.
Main Methods:
- A two-stage deep learning approach was employed, utilizing YOLOv5 and YOLOv8 for lesion detection.
- An ensemble method was explored to improve lesion detection performance.
- ProtoPNet, a transparent neural network, was used for classifying detected lesions into benign or malignant categories.
Main Results:
- The AI system achieved a recall of 0.76 for detection and an accuracy of 0.70 for classification.
- Expert radiologists provided positive clinical feedback, validating the system's potential clinical relevance.
- Dataset limitations and annotation accuracy challenges were noted, impacting final metric values.
Conclusions:
- The developed AI system offers a significant advancement in the timely and accurate analysis of DBT scans.
- Transparency and interpretability are crucial for AI systems in medical diagnostics, particularly in breast cancer screening.
- This research paves the way for improved early breast cancer detection and patient outcomes through explainable AI.
