Deep Learning Image Quality for Prostate MRI

About
Summary
This algorithm is a deep learning classifier designed to automatically assess the quality of axial T2-weighted prostate MRI scans. It was developed using a dataset of 1,229 examinations from a single institution, with quality labels provided by expert readers. The model was trained to distinguish between low- and high-quality images and to provide a continuous score reflecting overall image quality. To enhance interpretability, the algorithm also generates class-specific heatmaps that visualize which regions influenced the prediction.
Mechanism
Target population: Men undergoing prostate MRI for prostate cancer detection or evaluation, specifically focusing on axial T2-weighted scans.
Algorithm description: The method is based on a 3D convolutional neural network architecture (Inception I3D) adapted for prostate MRI quality classification. The network outputs a continuous score between 0 (poor quality) and 1 (excellent quality), alongside voxel-level Grad-CAM heatmaps highlighting regions that contributed to the decision. Both high-quality and low-quality attention maps are generated to provide interpretable feedback.
Inputs and outputs: • Input: One axial T2-weighted prostate MRI volume in .mha format. • Outputs: • A JSON file containing the continuous quality score (float between 0–1). • A heatmap for the low-quality class in .mha format. • A heatmap for the high-quality class in .mha format.
Interfaces
This algorithm implements all of the following input-output combinations:
Inputs | Outputs | |
---|---|---|
1 |
Validation and Performance
Uses and Directions
This algorithm was developed for research purposes only.