Deep-Learning Models for Detection and Localization of Visible Clinically Significant Prostate Cancer on...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Oncology Diagnostics

Background:

• Deep learning (DL) models are being explored for diagnosing clinically significant prostate cancer (csPCa).
• Further evaluation is needed for DL models in patients with prostate-specific antigen (PSA) levels between 4-10 ng/mL.

Purpose of the Study:

• To assess the efficacy of deep learning models using diffusion-weighted imaging (DWI) alone and in combination with T2-weighted imaging (T2WI).
• To detect and localize visible csPCa in patients with intermediate PSA levels.

Main Methods:

• A retrospective study involving 1628 patients with biopsy-confirmed csPCa or non-csPCa.
• Development of a diffusion model (DWI, apparent diffusion coefficient [ADC]) and a biparametric model (DWI, ADC, T2WI) using U-Net architecture.
• Performance evaluation based on lesion-level, location-level, and patient-level analyses using sensitivity, specificity, and area under the ROC curve (AUC).

Main Results:

• The diffusion model achieved high lesion-level sensitivity (89.0%), comparable to PI-RADS (90.8%).
• At the patient level, the diffusion model showed significantly higher sensitivity (96.0%) than the biparametric model (90.0%).
• Both DL models demonstrated significantly higher AUCs than PI-RADS assessment for location analysis.

Conclusions:

• The diffusion-based deep learning model is effective for detecting and localizing csPCa in patients with PSA levels of 4-10 ng/mL.
• Deep learning models, particularly the diffusion model, offer improved sensitivity for csPCa diagnosis in this patient group.