Detecting schizophrenia with AI

 From Nature.com.

Here is the link.

Here are some excerpts.

-----------------------------------------

Detecting schizophrenia with 3D structural brain MRI using deep learning

Abstract

Schizophrenia is a chronic neuropsychiatric disorder that causes distinct structural alterations within the brain. We hypothesize that deep learning applied to a structural neuroimaging dataset could detect disease-related alteration and improve classification and diagnostic accuracy. We tested this hypothesis using a single, widely available, and conventional T1-weighted MRI scan, from which we extracted the 3D whole-brain structure using standard post-processing methods. A deep learning model was then developed, optimized, and evaluated on three open datasets with T1-weighted MRI scans of patients with schizophrenia. Our proposed model outperformed the benchmark model, which was also trained with structural MR images using a 3D CNN architecture. Our model is capable of almost perfectly (area under the ROC curve = 0.987) distinguishing schizophrenia patients from healthy controls on unseen structural MRI scans. Regional analysis localized subcortical regions and ventricles as the most predictive brain regions. Subcortical structures serve a pivotal role in cognitive, affective, and social functions in humans, and structural abnormalities of these regions have been associated with schizophrenia. Our finding corroborates that schizophrenia is associated with widespread alterations in subcortical brain structure and the subcortical structural information provides prominent features in diagnostic classification. Together, these results further demonstrate the potential of deep learning to improve schizophrenia diagnosis and identify its structural neuroimaging signatures from a single, standard T1-weighted brain MRI.

Introduction

Schizophrenia is a progressive neuropsychiatric disorder that is characterized by structural changes within the brain. Recent findings from a large meta-analysis suggest that schizophrenia is associated with gray matter reductions across multiple subcortical regions including the hippocampus, amygdala, caudate, and thalamus, with structural changes in shape within those regions supporting changes in functional brain networks1. In addition to the altered shape of such brain structures, schizophrenia is also associated with significantly greater mean volume variability of the temporal cortex, thalamus, putamen, and third ventricle2. Other studies also affirm the enlargement of ventricles in schizophrenia3,4. While gray matter reductions are most consistently reported in the subcortical regions, reductions have also been identified in areas such as the prefrontal, temporal, cingulate, and cerebellar cortices5,6. Loss of gray matter volume has been shown to not only mark the onset of schizophrenia but also progress alongside the illness7.

Despite these documented changes, accurate and rapid detection of schizophrenia remains a pressing problem; previous studies are limited to only characterizing structural abnormalities at a group level, with no concrete method to make individual diagnoses at a subject level. Additionally, the diagnosis of schizophrenia based on DSM-5 criteria is costly both in terms of time and resources, without ensuring objectivity. Therefore, it is imperative to develop an objective screening tool to diagnose schizophrenia and potentially improve patient prognosis by allowing for earlier intervention.

Various attempts have been proposed to take advantage of the structural alterations present in schizophrenia for classification using neuroimaging data. Machine learning algorithms have historically presented the ability to classify psychiatric disorders in this manner8,9. In particular, the support vector machine (SVM), a supervised learning algorithm able to capture non-linear patterns in high-dimensional data, has been most prevalent in schizophrenia classification. Other popular machine learning algorithms for schizophrenia classification include multivariate pattern analysis, linear discriminant analysis, and random forest8,10. While standard machine learning approaches have demonstrated compelling results, their performance highly depends on the validity of manually extracted features8. Such features are traditionally extracted based on a combination of previously known disease characteristics and automatic feature selection algorithms11. These features may not completely encode the subtle neurological differences associated with schizophrenia; alternatively, they may encode too much unnecessary information requiring additional feature reduction12.

Deep learning has recently emerged as a new approach demonstrating superior performance over standard machine learning algorithms to classify neurological diseases using structural MRI data. Specifically, Convolutional Neural Networks (CNNs) can learn and encode the significant features necessary for classification and have become popular in medical image analysis13,14,15. This property makes CNNs uniquely suited to tasks like schizophrenia classification, where the specific features selected can dramatically impact model performance. Some studies have already demonstrated the utility of CNNs for schizophrenia classification. Oh J et al.16 achieved an impressive state-of-the-art performance (area under the ROC curve = 0.96) using 3D CNN for schizophrenia classification based on structural MRI data and was thus compared to as the benchmark model. Nevertheless, they struggled to generalize well on an unseen private dataset. Their inconsistent performance may be attributed to the dataset and patient variability as well as certain pre-processing choices, such as the inclusion of whole head as opposed to whole-brain MRI data and severe downsampling. Moreover, their region of interest analysis was limited and did not investigate brain structures in depth to inform specific changes in structural features associated with schizophrenia. Hu et al. combined structural and diffusion MRI scans for schizophrenia classification and found that 3D CNN models could outperform 2D pre-trained CNN models as well as multiple standard machine learning algorithms like SVM. Despite this, their best 3D model only reached the area under the ROC curve of 0.8417. As a consequence, though deep learning has advanced neuroimaging-based schizophrenia classification, the preprocessing and acquisition of large datasets coupled with the achievement of high model performance and generalization remains a great challenge.

In this study, we not only address the limitations in schizophrenia classification with T1-weighted (T1W) MRI data but also take advantage of class activation maps (CAM) in a deep learning network to visualize informative regions with disease vulnerability. Our main contributions include the following: firstly, we develop a 3D CNN using structural MRI scans to yield a performance better than the benchmark model16 for schizophrenia classification; and secondly, we apply gradient class activation maps to localize the brain regions related to schizophrenia identification. By visualizing feature activations, we provide further evidence that the structure of subcortical regions and ventricular areas1,2 are affected in schizophrenia.