Improved breast lesion detection in mammogram images using a deep neural network

Area of Science:

• Artificial Intelligence in Medical Imaging
• Machine Learning for Cancer Diagnosis
• Radiology and Breast Imaging Analytics

Background:

• Accurate and efficient breast cancer (BC) detection is crucial for patient outcomes.
• Radiologist interpretation of mammograms can be time-consuming and subject to variability.
• Deep neural networks (DNNs) show potential for augmenting diagnostic capabilities in medical imaging.

Purpose of the Study:

• To evaluate the impact of a deep neural network (DNN) on the accuracy of breast cancer detection.
• To assess the effect of a DNN on the time required for mammogram interpretation by radiologists.
• To compare the diagnostic performance of senior and junior radiologists with and without DNN assistance.

Main Methods:

• A retrospective study utilizing 880 mammograms from 220 patients.
• Development and application of a DNN model for detecting four malignancy features (masses, calcifications, asymmetries, architectural distortions).
• Comparison of radiologist performance (senior and junior) with and without DNN aid using Area Under the Curve (AUC) and assessment time.

Main Results:

• The DNN model achieved high AUCs for mass (0.877) and calcification (0.937) detection.
• DNN assistance significantly improved AUC values for mass, calcification, and asymmetric compaction detection in both senior and junior radiologists.
• The DNN model substantially reduced mammogram assessment time for both senior and junior radiologists.

Conclusions:

• The deep neural network demonstrates high accuracy in identifying key features of breast cancer on mammograms.
• Utilizing the DNN model led to a significant improvement in diagnostic accuracy for radiologists.
• The DNN effectively decreased the time needed for mammogram review, enhancing workflow efficiency.