A deep learning model, NAFNet, predicts adverse pathology and recurrence in prostate cancer using MRIs

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate prediction of adverse pathology and biochemical recurrence-free survival (bRFS) is crucial for prostate cancer management.
• Current risk stratification tools may have limitations in predicting patient outcomes.

Purpose of the Study:

• To apply the NAFNet deep learning network for predicting adverse pathology and bRFS in prostate cancer using pre-treatment MRI.
• To develop and validate a deep learning-based nomogram (DL-nomogram) for improved risk stratification.

Main Methods:

• Trained and validated the NAFNet model on MRI data from 514 prostate cancer patients across multiple centers.
• Constructed a DL-nomogram integrating NAFNet predictions with clinical T stage and biopsy results.
• Compared the performance of the DL-nomogram against PI-RADS and CAPRA scores.

Main Results:

• NAFNet-classifier outperformed ResNet50 in predicting adverse pathology in the external test set.
• The DL-nomogram achieved the highest AUC (0.915) and accuracy (0.850) for predicting adverse pathology.
• The DL-nomogram demonstrated superior performance (C-index: 0.732) compared to the CAPRA score in predicting bRFS.

Conclusions:

• The developed DL-nomogram, powered by NAFNet, accurately predicts adverse pathology and prognosis in prostate cancer.
• This AI-driven tool shows potential for enhancing risk stratification in medical imaging for prostate cancer.