Development and validation of a deep learning-based automatic classification algorithm for the medial temporal lobe...

The aim of this study was to develop and validate a deep learning-based automatic classification algorithm for the medial temporal lobe atrophy (MTA) score in patients with cognitive impairment.

This retrospective, observational study included consecutive patients with cognitive impairment from a tertiary hospital between March 2017 and June 2021. We developed a deep learning-based model and a machine learning-based model to automate MTA classification. We reorganised the MTA scores into 3 classes (0/1), (2), and (3/4) then classified the right and left MTA scores separately. The internal testing and external testing datasets were applied and compared to validate the performance of the MTA prediction model.

A total of 1694 patients were evaluated for the training dataset, and 297 patients evaluated for the internal testing dataset. 400 patients were evaluated for the external testing dataset. In the internal testing dataset, the accuracy was 0.82 and 0.87 for the left and right MTA classifications, respectively. In the external testing dataset, the accuracy was 0.82 and 0.85 for the left and right MTA classifications, respectively. When comparing the performance between a deep learning-based model and a machine learning-based model, the results were similar.

The deep learning- and machine learning-based automatic classification algorithms for the MTA score accurately classified the MTA score in patients with cognitive impairment.