Combining Clinical Data With Neuro Images to Identify the Treatment Resistant in Depression by NLP

Abstract

Predicting treatment-resistant depression (TRD) is difficult, even though 21% of individuals with depression who get therapy do not achieve remission. The purpose of this research is to use structured data from electronic health records, brain morphology, & natural language processing to create a multimodal forecast model for TRD that can be explained. A total of 248 patients who recently had a period of depression were included. Combining topic probability from clinical notes with separate components-map weights from brain T1-weighted MRI, and chose tabular dataset attributes, TRD-predictive models were created. All of the models used five-fold cross-validation to apply the XGBoost algorithm. The area under the receiver's operating characteristic was 0.795 for the model that utilized all data sources, then for models that used structured data and brain MRI together, and finally for models that used brain MRI and medical records separately. (0.771), (0.763) plus structured data, (0.729) plus clinical notes, (0.704) plus structured data,