Prediction and Mapping of Intraprostatic Tumor Extent with Artificial Intelligence

Area of Science:

• Urology
• Medical Imaging
• Artificial Intelligence

Background:

• Magnetic resonance imaging (MRI) often underestimates prostate cancer extent, complicating focal treatment margin definition.
• Accurate margin definition is crucial for effective focal therapy and reducing recurrence.

Purpose of the Study:

• To validate focal treatment margins generated by an artificial intelligence (AI) deep learning model.
• To compare the accuracy of AI-defined margins with conventional methods.

Main Methods:

• A retrospective analysis of 50 radical prostatectomy cases for intermediate-risk prostate cancer.
• An AI model integrated multimodal imaging and biopsy data to create 3D cancer maps and margins.
• AI margins were compared against conventional MRI regions of interest (ROIs), 10-mm ROI margins, and hemigland margins.

Main Results:

• AI margins demonstrated significantly higher sensitivity (97%) for detecting cancer-bearing voxels compared to conventional methods (37%-94%, p < 0.001).
• AI margins resulted in more negative margins (90%) for index lesions than conventional methods (0%-66%, p < 0.004).
• The AI model accurately predicted negative surgical margin probability (R² = 0.98).

Conclusions:

• The AI model proved accurate and effective in an independent test set for defining prostate cancer margins.
• This AI approach has the potential to standardize margin definition, improve focal treatment outcomes, and reduce cancer recurrence.
• Accurate margin prediction can aid informed decision-making for patients and clinicians.