Comprehensive Summary
With an Artificial Intelligence (AI)- based deep learning model to classify optical coherence tomography (OCT) images, it is possible to distinguish between control patients and patients with schizophrenia (SZ) using retinal biomarkers. This study utilizes Self-Attention NeXt, a convolutional neural network composed of inverted bottleneck blocks, integrated self-attention mechanisms, and a 1x1 convolution. To ensure the model's functionality, it was trained and tested on a publicly available dataset of OCT images collected from patients with SZ in 2017 (OCT2017). The AI Self-AttentionNeXt achieved a 97.0% accuracy on the collected SZ OCT dataset and 95% on the public OCT2017 dataset. The mapped visualizations of the gradient-weighted class activation further confirmed the model’s ability to attend to retinal regions and localize features to classify OCT images accurately.
Outcomes and Implications
With the implication of the Self-Attention NeXt, a combination of attention mechanisms with neural network interfaces can be used to detect and support patients with SZ earlier than otherwise, using OCT images. The use of this deep learning model can provide more thorough support for clinical decision-making and more accurate psychiatric diagnoses.