157. Validity verification of machine vision combined with biosensors IN the early screening of bipolar disorder

Abstract Background Bipolar disorder, as a type of mental illness with a high rate of misdiagnosis and recurrence, its early symptoms are often confused with unipolar depression. Moreover, clinical screening mostly relies on subjective scales, which leads to a delay in the timing of disease intervention. In existing research, machine vision technology mostly focuses on the recognition of patients' facial expressions or eye movement features, but it is easily disturbed by the environment. Meanwhile, biosensors have problems such as large individual differences and insufficient specificity, making it difficult to distinguish them from other emotional disorders. Therefore, the research aims to integrate machine vision technology with biosensors to design a multimodal fusion early screening program, in order to enhance the accuracy of early screening for bipolar disorder. Methods A total of 216 subjects were included in the study, including 108 patients diagnosed with early bipolar disorder (BD group) and 108 healthy controls (HC group). There were no significant differences in baseline data such as age and gender among the three groups of participants. The study collected 15 facial expression features and 8 eye movement parameters of the subjects through machine vision technology, and used portable biosensors to monitor the levels of Cortisol (COR) in saliva and related indicators of Heart Rate Variability (HRV) of the participants. Subsequently, the random forest algorithm was used to construct a multimodal joint screening model, and the independent sample t-test was adopted to compare the differences of indicators between groups to verify the inter-group discrimination of the core predictive indicators of the model. The main indicators include Positive Predictive Value (PPV), Negative Predictive Value (NPV), and Area Under the ROC Curve (AUC), etc. Results The overall screening effect of the combined plan was excellent, with an AUC value as high as 0.942, a sensitivity of 90.75%, and PPV and NPV being 91.60% and 91.83% respectively. The comparison results of the predictive indicators between the BD group and the HC group are shown in Table 1. It can be seen from Table 1 that the eye fixation duration (2.07 ± 0.41 s) and HRV triangle index (11.95 ± 2.06) in the BD group were significantly lower than those in the HC group, and the differences were statistically significant (p<.001). However, the frequency of frowning (8.78 ± 2.34), COR (19.26 ± 3.17 ng/mL), and the predicted probability (87.95 ± 9.8%) were much higher than those in the HC group, and the differences between the groups were also statistically significant (p<.001). Discussion Research has demonstrated that the combined application of machine vision and biosensors not only avoids the problem of environmental interference caused by machine vision but also makes up for the deficiency of specificity of biosensors, which can provide an objective and effective technical solution for the early screening of bipolar disorder.