Precise Identification of Prostate Cancer from DWI Using Transfer Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence in Oncology
• Radiology

Background:

• Computer-aided detection (CAD) systems enhance radiologist objectivity and reduce reliance on invasive procedures.
• Prostate cancer detection can be improved through advanced imaging analysis.
• Diffusion-weighted imaging (DWI) offers valuable insights into tissue characteristics.

Purpose of the Study:

• To develop and evaluate a CAD system for detecting and identifying prostate cancer using DWI.
• To assess the performance of deep learning models in analyzing DWI data for prostate cancer diagnosis.

Main Methods:

• Non-negative matrix factorization (NMF) for prostate region segmentation.
• Apparent diffusion coefficient (ADC) volume estimation and radiologist-based labeling.
• Transfer learning with fine-tuned convolutional neural network (CNN) models (AlexNet, VGGNet) for classification.

Main Results:

• Experiments evaluated CNN models on DWI datasets with nine b-values.
• AlexNet achieved an average accuracy of 89.2±1.5%, with sensitivity 87.5±2.3% and specificity 90.9±1.9%.
• VGGNet demonstrated improved performance with an average accuracy of 91.2±1.3%, sensitivity 91.7±1.7%, and specificity 90.1±2.8%.

Conclusions:

• The developed CAD system is feasible and accurate for prostate cancer detection using DWI.
• Deeper CNN models like VGGNet significantly improve detection accuracy.
• The system shows promise for enhancing non-invasive prostate cancer diagnosis.