Assessment of Prostate Cancer Aggressiveness Through the Combined Analysis of Prostate MRI and 2.5D Deep Learning Models

Yalei Wang¹Yuqing Xin²Fuqiang Pan²Baoqi Zhang¹Xu Li²Manman Zhang²Yushan Yuan³Lei Zhang³Peiqi Ma³Bo Guan^4*Yang Zhang^1*

• ¹Department of Radiology, Affiliated Fuyang People's Hospital of Anhui Medical University, Fuyang, China
• ²Department of Radology, Affiliated Fuyang People's Hospital of Bengbu Medical University, Fuyang, Anhui, China, Fuyang, China
• ³Fuyang City People's Hospital, Fuyang, Anhui Province, China
• ⁴Department of Urology, Fuyang City People's Hospital, Fuyang, Jiangsu Province, China

Objective: Prostate cancer is prevalent among older men. Although this malignancy has a relatively low mortality rate, its aggressiveness is critical in determining patient prognosis and treatment options. This study therefore aimed to evaluate the effectiveness of a 2.5D deep learning model based on prostate MRI to assess prostate cancer aggressiveness.This retrospective study included 335 prostate cancer patients (266 aggressive, 69 non-aggressive) confirmed by pathology between January 2022 and December 2023. All underwent biparametric MRI—T2-weighted, diffusion-weighted, and ADC sequences—prior to biopsy. Patients were randomly divided into training and test sets (7:3).Lesions were manually segmented by two radiologists using ITK-SNAP. The largest lesion slice and adjacent slices above and below formed 2.5D inputs. Radiomic features were extracted via Pyradiomics, and deep learning features were derived using the Inception_v3 network. Only features with good interobserver consistency (ICC > 0.75) were retained, normalized, and filtered using t-tests, Pearson correlation, and Lasso regression.Four LightGBM models were built: radiomic (Rad-LightGBM), deep learning (DL-LightGBM), combined (DLR-LightGBM), and clinical data (Clinic-LightGBM). The best feature model was combined with clinical variables to develop a nomogram. Grad-CAM and LightGBM visualization were used to interpret model decisions. Performance was evaluated by AUC, sensitivity, and specificity. Results: In the test set, the nomogram demonstrated the highest predictive ability for prostate cancer aggressiveness (AUC = 0.919, 95% CI: 0.8107-1.0000), with a sensitivity of 0.966 and specificity of 0.833. The DLR-LightGBM model (AUC = 0.872) outperformed the DL-LightGBM (AUC = 0.818) and Rad-LightGBM (AUC = 0.758) models, indicating the benefit of combining deep learning and radiomic features. Conclusion: Our 2.5D deep learning model based on prostate MRI showed efficacy in identifying clinically significant prostate cancer, providing valuable references for clinical treatment and enhancing patient net benefit.