Detecting Schizophrenia by Structural MRI Using Deep Learning

Abstract

Schizophrenia affects about 1% of the world population with a lifetime prevalence of 0.3- 0.66% is among the frequently occurring psychotic disorders. Schizophrenia has been clinically identified, but there is no pathophysiology for diagnosing it. Despite the fact that much study has been done on volumetric MRI in Schizophrenia, detecting it using biological markers is difficult as most psychiatric diseases share the same symptoms. The prefrontal and temporal lobes, particularly the medial and superior temporal lobes, have been found to have decreased volume in earlier research. Predictive analytics and clinical decision assistance both heavily rely on these findings. Analyzing enormous medical imaging data is time consuming for experts. Additionally, drawing inferences from these analyses may be erroneous or biased. While machine learning algorithms may aid specialists in automated analysis to some level, they might not be able to analyze such vast volumes of data and accurately resolve complex problems as these conventional methods often overlook important information. We suggest employing a deep learning approach based on the Convolutional Neural Networks (CNN) rather than the conventional machine learning approaches. Deeper networks can learn a new, more complex representation of the input data at each layer, allowing them to develop deep representations. Clinicians may find it easier to distinguish schizophrenia from other mental diseases using deep learning algorithms that have been trained on bigger data sets with a variety of illness stages and severity.