Exploring Uncertainty Measures in Bayesian Deep Attentive Neural Networks for Prostate Zonal Segmentation

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate segmentation of prostatic zones on multiparametric MRI (mpMRI) is crucial for improving prostate cancer diagnosis.
• Current segmentation methods can be time-consuming and may lack precision.

Purpose of the Study:

• To develop and validate a spatial attentive Bayesian deep learning network for automatic segmentation of prostate peripheral zone (PZ) and transition zone (TZ).
• To assess the uncertainty estimation of the proposed model in segmenting these zones.

Main Methods:

• A spatial attentive Bayesian deep learning network was designed for PZ and TZ segmentation.
• The model was trained and validated on internal datasets (PROSTATEX) and tested on an independent external dataset (ETD).
• Segmentation performance was quantified using the Dice Similarity Coefficient (DSC).

Main Results:

• The proposed deep learning method achieved high mean DSCs for PZ (0.80±0.05 internal, 0.79±0.06 external) and TZ (0.89±0.04 internal, 0.87±0.07 external).
• Performance was consistent across internal and external datasets, with no significant differences.
• The method outperformed state-of-the-art segmentation techniques.

Conclusions:

• The developed deep learning model accurately segments prostate PZ and TZ with reliable uncertainty estimation.
• This automated approach enhances diagnostic workflow for prostate cancer.
• Segmentation uncertainty was found to be highest at the junction of PZ, TZ, and AFS, correlating with model performance.