Prostate cancer (PCa) is one of the leading cancer types in terms of new cancer cases and deaths in men worldwide [1]. However, it presents various forms of aggressiveness, contributing to the challenge of over diagnosing indolent disease. Properly managing PCa requires accurately assessing its presence and severity to avoid misjudging patients’ conditions [2].

The Gleason score based on invasive biopsies is the grading system for determining PCa aggressiveness. This score provides important prognostic information for patients and is an essential element for treatment planning [3]. Noninvasive approaches for detecting PCa and distinguishing indolent from deadly disease on MRI might assist in early diagnosis and treatment planning. Multiparametric magnetic resonance imaging (mpMRI) may aid in PCa localization and targeted prostate biopsies. However, detecting clinically significant PCa (csPCa) is a major challenge. There is increasing evidence suggesting the high accuracy of mpMRI in ruling out clinically significant diseases. A priority in patient management is the accurate classification of csPCa to assess the risk of malignant progression and metastasis, thereby avoiding overtreatment in low-risk patients and undertreatment in high-risk patients choosing active surveillance.

Prostate-specific antigen (PSA) is a commonly used clinical biomarker for the screening and early diagnosis of PCa. However, its application as a PCa biomarker has raised concerns regarding overdiagnosis and overtreatment of indolent disease [4]. In clinical practice, prostate MRI was initially employed for tumor staging before radical prostatectomy or radiation therapy. mpMRI technology combines T2-weighted (T2W) sequences and functional MRI sequences such as diffusion-weighted imaging (DWI), apparent diffusion coefficient (ADC), and dynamic contrast-enhanced MRI, thereby improving disease detection, staging, risk stratification, and treatment planning for PCa [5]. The Prostate Imaging Reporting and Data System (PI-RADS) has standardized PCa diagnosis using MRI [6]. However, MRI is still restricted by benign confounding appearances and substantial intrareader and inter-reader variability. More robust quantification methods are essential to improving diagnostic and staging accuracy and refining surveillance strategies. mpMRI offers increasing accuracy in detecting csPCa, and PI-RADS has been widely used to characterize and grade focal intraglandular prostate nodules, leading to limited diagnostic accuracy and interpretation inconsistencies among radiologists for detecting csPCa.

Deep learning (DL) involves the use of computational algorithms that can make accurate discriminations without explicit pre-instructions. Integrating DL with mpMRI represents a cutting-edge solution for PCa characterization [7,8]. This study primarily aimed to develop an automatic mpMRI tool for identifying malignant lesions in the prostate and classifying csPCa in suspicious patients. In this study, we trained two different DL models, with pathological and clinical examinations serving as the ground truth. The sensitivity and specificity of the DL-mpMRI system were compared with PSA and PI-RADS in csPCa detection. The integration of DL models with mpMRI marks a significant advancement in assessing PI-RADS and aids in efficiently identifying csPCa lesions.