Related Experiment Video
Updated: May 10, 2026
08:05
Detection and Isolation of Cancer in Prostate Biopsies Using Stimulated Raman Histology and Artificial Intelligence
Published on: June 10, 2025
Prostate Cancer Detection on Micro-Ultrasound Raw Data Using a Deep Learning Neural Network
Ahmed El Kaffas1, Thodsawit Tiyarattanachai2, Mirabela Rusu2
1Department of Radiology, UCSD, La Jolla, CA, USA.
Ultrasound in Medicine & Biology
|May 8, 2026
Summary
A new deep learning model, PSNet, analyzes micro-ultrasound power spectrums to detect prostate cancer. Combining PSNet with PSA measurements significantly improves diagnostic accuracy for clinically significant prostate cancer.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Micro-ultrasound (micro-US) offers high-resolution imaging for prostate biopsies.
- Interpreting micro-US images during live biopsies presents a significant challenge.
Purpose of the Study:
- Develop a convolutional neural network (CNN) to classify prostate tissues from micro-US power spectrums (PS).
- Distinguish between benign tissues and clinically significant prostate cancer (csPCa).
- Create a tool for automated interpretation during image-guided prostate biopsies.
Main Methods:
- A custom CNN, PSNet, was developed using retrospective micro-US data from 491 men.
- PS were derived from raw micro-US data, avoiding prostate segmentation.
- Model performance was evaluated against traditional CNNs using B-mode images, with histopathology as the ground truth.
Main Results:
- PSNet achieved an ROC-AUC of 82% for classifying benign vs. csPCa from PS alone.
- Incorporating prostate-specific antigen (PSA) improved ROC-AUC to 85% at the frame level.
- At the patient level, PSNet with PSA achieved a high ROC-AUC of 91% and specificity of 99%.
Conclusions:
- Deep learning can identify prostate cancer from micro-US acoustic properties without segmentation.
- PSA measurements enhance the diagnostic value of micro-US data, particularly specificity.
- This approach shows potential for guiding targeted prostate biopsies more effectively.