Auxiliary diagnosis models of bipolar disorder based on functional magnetic resonance imaging and deep learning

魏鑫茹,段佳,张然,杨景钰,张陆衡,姚菲,董帅,张锡哲,王菲,朱荣鑫

DOI: https://doi.org/10.3760/cma.j.cn113661-20210430-00147

2022-01-01

Abstract:Objective:Construction of deep learning classification models based on functional magnetic resonance imaging (fMRI) data assists the clinicians to achieve better diagnosis of bipolar disorder (BD), which can improve the recognition rate of BD by identifying the critical imaging features.Methods:A total of 146 patients who met the diagnosis criteria of BD according to DSM-Ⅳ and 234 healthy control (HC) were recruited for fMRI scans. Regional homogeneity (ReHo) and amplitude of low frequency fluctuation (ALFF) were used to analyze fMRI data. Based on ReHo and ALFF, the classification models were constructed by deeping neural network (DNN) and dual-channel convolution neural network (DCNN) respectively, and the best classification model was developed by comparing the accuracy and area under curve (AUC) of the two models. Based on each brain region divided by anatomical automatic labeling (AAL), the support vector machine (SVM) classification model was constructed using imaging index with a better performance, and the critical imaging features were identified by comparing the accuracy of each brain region.Results:The performances of the DCNN classification model (accuracy = 75.3%, and 72.6%, respectively, based on ReHo and ALFF) were significantly better than the DNN classification model (accuracy = 67.1%, and 65.1%, respectively). Meanwhile, the accuracy of classification model constructed using ReHo was higher than ALFF. Based on the SVM classification model, critical brain regions were identified above the accuracy of 65.0%, including the occipital lobe (middle occipital gyrus, superior occipital gyrus and lingual gyrus), hippocampus, and thalamus.Conclusion:The computational model based on DCNN using ReHo can help the clinicians to achieve better diagnosis of BD. Furthermore, occipital lobe, hippocampus and thalamus may be the critical imaging features for the auxiliary recognition of BD.