Improving Benign and Malignant Classifications in Mammography with ROI-Stratified Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Deep learning is widely adopted for medical image diagnosis, particularly in mammography for breast cancer screening.
• Accurate differentiation between benign and malignant breast lesions is crucial for effective screening and patient management.

Purpose of the Study:

• To investigate if incorporating region-of-interest (ROI) mask information improves deep learning model accuracy for benign/malignant mammographic diagnoses.
• To evaluate the performance of Swin Transformer and ConvNeXtV2 models with and without ROI mask integration.

Main Methods:

• Utilized Swin Transformer and ConvNeXtV2 models on the VinDr and CDD-CESM mammography datasets.
• Stratified images based on the presence or absence of ROI masks, training and predicting independently for each subgroup.
• Merged results from ROI-stratified subgroups to assess overall diagnostic performance.

Main Results:

• ROI-stratified analysis significantly improved prediction metrics (sensitivity, specificity, F-score, accuracy) compared to baseline.
• For instance, VinDr/Swin Transformer improved from (0.00, 1.00, 0.00, 0.85) to (0.93, 0.87, 0.90, 0.87) with ROI masks.
• CDD-CESM/ConvNeXtV2 showed improvement from (0.65, 0.65, 0.65, 0.65) to (0.74, 0.61, 0.67, 0.68) with ROI masks.

Conclusions:

• The study validates the hypothesis that incorporating ROI mask information enhances deep learning model performance in mammography.
• Considering ROI mask data is crucial for improving diagnostic accuracy in distinguishing benign from malignant breast lesions.
• This approach offers a promising strategy for more precise and reliable AI-assisted mammographic interpretation.