Automated Detection and Grading of Extraprostatic Extension of Prostate Cancer at MRI via Cascaded Deep Learning and...

Area of Science:

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

Background:

• Extraprostatic extension (EPE) is a key indicator of prostate cancer aggressiveness and recurrence risk.
• Accurate pre-operative assessment of EPE is crucial for planning radical prostatectomy and surgical approach.
• Current methods for EPE assessment can be subjective and vary in accuracy.

Purpose of the Study:

• To develop and evaluate a deep learning-based artificial intelligence (AI) workflow for automated grading of EPE from prostate MRI.
• To assess the performance of the AI workflow in predicting EPE compared to histopathology.
• To explore the potential of AI in improving the consistency and accuracy of EPE assessment.

Main Methods:

• A cohort of 634 patients with prospective MRI assessments was used, with a training set of 507 and a test set of 127 patients.
• Deep learning models for prostate segmentation were utilized to extract features for random forest classification.
• Model performance was evaluated using balanced accuracy, ROC AUCs, sensitivity, specificity, and accuracy against histopathology ground truth.

Main Results:

• The AI workflow achieved a balanced accuracy of 0.390 ± 0.078.
• ROC AUCs for AI-assigned EPE grades 0-3 were 0.70, 0.65, 0.68, and 0.55, respectively.
• For EPE ≥ 1, the AI model achieved 0.67 sensitivity, 0.73 specificity, and 0.72 accuracy, compared to radiologist performance of 0.81 sensitivity, 0.62 specificity, and 0.66 accuracy.

Conclusions:

• The developed AI workflow demonstrates accuracy in predicting histologic EPE that approaches physician performance.
• This automated system offers potential for enhancing physician decision-making in assessing EPE risk for prostate cancer patients.
• The AI workflow's consistency and automation can improve the reliability of EPE grading in clinical practice.