Multimodal imaging deep learning model for predicting extraprostatic extension in prostate cancer using MpMRI and...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate prediction of extraprostatic extension (EPE) is crucial for prostate cancer (PCa) treatment planning.
• Current diagnostic methods have limitations in precisely determining EPE.
• Multimodal imaging offers potential for improved diagnostic accuracy.

Purpose of the Study:

• To develop a multimodal deep learning (DL) model integrating magnetic resonance imaging (mpMRI) and 18F-PSMA-PET/CT for EPE prediction in PCa.
• To evaluate the DL model's effectiveness in enhancing radiologist diagnostic accuracy for EPE.
• To assess the clinical net benefit of the developed DL model.

Main Methods:

• Retrospective collection of clinical and imaging data from PCa patients who underwent radical prostatectomy.
• Development of a multimodal DL model combining mpMRI and 18F-PSMA-PET/CT.
• Comparison of the multimodal DL model's performance against single-modality models and radiologist assessments with and without model assistance.

Main Results:

• The combined mpMRI + PET/CT multimodal DL model achieved an AUC of 0.82 in the primary institution and 0.81 in the external validation set.
• The multimodal DL model demonstrated superior predictive accuracy compared to single-modality models.
• Deep learning-assisted assessment significantly improved AUC and sensitivity (P < 0.05) compared to radiologists alone, with a modest specificity reduction.

Conclusions:

• A multimodal deep learning model integrating mpMRI and 18F-PSMA-PET/CT shows strong predictive performance for EPE in prostate cancer.
• The model effectively enhances radiologist accuracy in EPE assessment.
• This AI tool shows potential for supporting individualized and precise therapeutic decision-making in prostate cancer management.