Related Experiment Video
Updated: Mar 18, 2026
A Cognitive Fusion-guided Prostate Biopsy Using Multiparametric Magnetic Resonance Imaging and Transrectal Ultrasound
Published on: March 21, 2025
PRAD-Hybrid CNN (PRADHC): A Deep Learning Model for Assisted Diagnosis of Prostate Cancer on MRI
Jingpeng Liu1, Lingxuan Hou2, Yang Xu3
1Department of Urology, Beijing Tiantan Hospital, Capital Medical University, Beijing, China.
Background:
Prostate cancer is a prevalent malignancy in males, with prostate MRI imaging as the primary diagnostic method. However, this method is subjective and can miss early-stage cancers, necessitating more efficient diagnostic techniques.
Method:
In this study, we introduced the PRAD-Hybrid CNN (Prostate Adenocarcinoma Hybrid Convolutional Neural Network, PRADHC) model, a novel amalgamation of EfficientNet and Residual Blocks, which was developed and validated on 1,528 MRI images from 64 patients. By strategically increasing the number of Convolutional Neural Network (CNN) layers in the EfficientNet architecture, our model improved the diagnostic accuracy inherent to the original EfficientNet. Additionally, the integration of Residual Networks (ResNet) successfully mitigated the gradient vanishing issue often encountered during the training of deeper models, thereby significantly enhancing training accuracy. This innovative model, thus, offers clinicians an efficacious tool for assisted diagnosis.
Result:
The PRADHC model, upon validation, achieved an accuracy of 99.34% and an AUC of 99.34%, a 4% improvement over the conventional EfficientNet. The baseline elementary CNN model achieved 95.72% accuracy and 96.74% AUC, which are still lower than the PRADHC model.
Discussion:
The superior performance of PRADHC can be attributed to the synergistic integration of EfficientNet’s multi-scale feature extraction and residual learning, which facilitates deeper network optimization without degradation. Compared with single-architecture CNN models, the hybrid design enhances robustness to MRI appearance variability and improves discrimination between malignant and non-significant prostate tissue
Conclusion:
This study introduces a novel deep learning model specifically designed for automated prostate cancer diagnosis. This model aims to enhance diagnostic accuracy, especially in the early stages of the disease. Such advancements have the potential to enhance the diagnostic proficiency of both radiologists and urologists, enabling more informed treatment planning. However, it is imperative to acknowledge that false-positive lesion detections remain a limitation of AI-assisted diagnostic tools. Nevertheless, this system can serve as a valuable supplementary instrument for radiologists in their diagnostic endeavors.
