Architectural Distortion-Based Digital Mammograms Classification Using Depth Wise Convolutional Neural Network.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Architectural distortion (AD) is a subtle mammographic finding, often challenging to detect, yet it is the third most suspicious sign of breast cancer.
• Accurate detection of AD is crucial for early diagnosis and improving patient outcomes in breast cancer screening.

Purpose of the Study:

• To develop and evaluate an automated computer-aided diagnostic system for detecting architectural distortion regions of interest (ROIs) in mammograms.
• To classify detected AD ROIs as malignant or benign using deep learning and machine learning techniques for breast cancer prediction.

Main Methods:

• A computer vision and deep learning framework was proposed, involving image preprocessing, augmentation, pixel-wise segmentation, and AD ROI detection.
• The system utilized depth-wise 2D V-net 64 convolutional neural networks for classification, trained and tested on PINUM, CBIS-DDSM, and DDSM mammogram databases.

Main Results:

• The proposed system achieved high classification accuracies of 0.95, 0.97, and 0.98 on the three evaluated mammogram databases.
• The method demonstrated superior performance compared to established models like ShuffelNet, MobileNet, SVM, K-NN, and RF, as well as previous studies.

Conclusions:

• The developed automated system effectively detects and classifies architectural distortions in mammograms, showing significant potential for enhancing breast cancer diagnosis.
• The computer vision and deep learning approach offers a promising tool to assist radiologists in identifying subtle ADs, potentially leading to earlier and more accurate breast cancer detection.