ProCUSNet: Prostate Cancer Detection on B-mode Transrectal Ultrasound Using Artificial Intelligence for Targeting...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Prostate Cancer Diagnostics

Background:

• Conventional B-mode transrectal ultrasound (TRUS) has limitations in detecting clinically significant prostate cancer.
• Accurate localization of cancerous lesions is crucial for effective biopsy and treatment planning.

Purpose of the Study:

• To evaluate the efficacy of an artificial intelligence (AI) method, ProCUSNet, in detecting clinically significant prostate cancer using B-mode transrectal ultrasound images.
• To compare the AI's performance against human readers and MRI interpretation.

Main Methods:

• Trained the PROstate Cancer detection on B-mode transrectal UltraSound images NETwork (ProCUSNet) using data from 2986 men across two institutions.
• ProCUSNet utilizes nnUNet frameworks for detecting and outlining cancer on 3D reconstructed ultrasound volumes.
• Performance was benchmarked against reference labels, eight ultrasound readers, and radiologists interpreting MRI.

Main Results:

• ProCUSNet identified 82% of clinically significant prostate cancer cases with high accuracy (lesion boundary error up to 2.67 mm).
• The AI detected 42% more lesions than human ultrasound readers (sensitivity 0.86 vs 0.44).
• ProCUSNet demonstrated comparable performance to MRI interpretation by radiologists (sensitivity 0.79 vs 0.78) and similar targeting utility as a supplement to biopsies.

Conclusions:

• ProCUSNet effectively localizes clinically significant prostate cancer on B-mode ultrasound, overcoming challenges faced by human readers.
• The AI shows potential as an adjunct to systematic biopsies for improved lesion targeting.
• ProCUSNet may offer a viable alternative for prostate cancer screening, potentially in the absence of MRI.