Depression detection from multimodalities based on LeNet with hunter-geese optimization.

Depression is considered a psychological illness, and the effects of depression are persistent sadness, loss of interest, and severe depression, which, if untreated, can lead to suicidal thoughts and actions. Moreover, early intervention and detection of depression are essential for reducing the loss of life and avoiding adverse complications. Conventional methods of diagnosing depression rely largely on clinician's judgment and the patient's self-reporting, which can sometimes result in variability or reduced accuracy in diagnoses. Approaches that rely on a single type of data often miss important aspects of depression, as they cannot fully represent the wide range of characteristics linked to the condition. To address these shortcomings, a novel framework named Hunter Geese Migration Optimization-based LeNet (HGMO-LeNet) is established for depression detection from multimodalities, comprising Magnetic Resonance Imaging (MRI) and speech. Primarily, the input MRI images are forwarded to the image processing, in which an adaptive Wiener filter is utilized to perform the image processing. Then, ROI extraction is carried out, which involves segmenting the brain region from the overall scan by removing non-brain tissues. Subsequently, feature extraction is done, and the features, namely volumetric and textural features, are extracted. Further, depression detection is established by exploiting the LeNet, which is fine-tuned by the designed HGMO. At the same time, the Gaussian filter assists in performing speech signal pre-processing. Further, the preprocessed signal is fed into the feature extraction phase, and then depression detection is accomplished using HGMO-LeNet. Finally, the output obtained from both inputs are merged using a correlation coefficient to produce the result. The devised HGMO-LeNet measured the maximum specificity of 91.000%, the accuracy of 91.276%, and the sensitivity of 92.266%. This demonstrates the superior performance of the devised model over existing methods and provides a robust system for automated depression detection.