Using deep learning to optimize the prostate MRI protocol by assessing the diagnostic efficacy of MRI sequences

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Biparametric prostate MRI (bpMRI) is crucial for prostate cancer diagnosis.
• Diffusion-weighted imaging (DWI) sequences are key components of bpMRI protocols.
• Optimizing MRI protocols can improve efficiency and patient throughput.

Purpose of the Study:

• To evaluate the diagnostic efficacy of different MRI sequences for prostate cancer detection.
• To explore the use of deep learning in optimizing prostate MRI protocols.
• To assess if accelerated MRI protocols maintain diagnostic performance compared to the full protocol.

Main Methods:

• A retrospective study of 840 patients with bpMRI scans.
• Simulated two accelerated MRI protocols by using fewer DWI b-values.
• Trained deep learning models to detect prostate cancer lesions on both accelerated and full protocols.
• Compared diagnostic performance using AUROC and pAUFROC metrics, validated by expert radiologists.

Main Results:

• No significant difference in diagnostic performance between accelerated and full bpMRI protocols.
• Omitting the b800 DWI sequence reduced scan time by 53% with minimal impact on AUROC (+0.01) and pAUFROC (-0.03).
• Omitting the b400 DWI sequence reduced scan time by 32% with negligible impact on AUROC (-0.01) and pAUFROC (+0.01).

Conclusions:

• Deep learning effectively assesses the diagnostic efficacy of MRI sequences for prostate MRI protocols.
• Optimizing prostate MRI protocols by omitting either b400 or b800 DWI sequences is feasible.
• Reduced scan times can be achieved without compromising diagnostic quality in prostate cancer detection.