Introduction

Rationale

Depression is a common mental health illness worldwide and a major contributor to the overall global burden of disease. It can cause the affected person to suffer greatly and function poorly at work, at school and in the family. At its worst, depression can lead to suicide. As one of the major urban health challenges, there are more than 300 millions people around the world suffering from depressive disorder, including 1.5 millions people in Thailand. Furthermore, the disability have a higher risk of becoming depressive, due to the limitation of the treatment accessibility and affordability [1].

Preventive healthcare service of this depressive disorder will help improve a quality of life and reduce a suicidal rate of the people in any society. However, such a solution is not widely accessible in Thailand, due to resource constraints and negative perspective of the mental disorder [2].

Artificial Intelligence (AI) technology has played a more significant role in the innovation of the medical industry. The practice of medicine is changing with the development of new AI methods of machine learning. Coupled with rapid improvements in computer processing, these AI-based systems are already improving the accuracy and efficiency of diagnosis and treatment across various specializations [3].

In the area of psychiatry, facial geometry and speech analysis for depression detection systems with the potential of serving as a decision support system was proposed, based on novel features extracted from facial expression geometry and speech, by interpreting non-verbal manifestations of depression. These findings suggest the feasibility of automatic detection of depression, raise new issues in automated facial image analysis and machine learning, and have exciting implications for clinical theory and practice [4-5].

Review Literature

Depression Severity Evaluation

Depression severity evaluation using a credible tool is essential for the depressive disorder detection, prevention and treatment (6). In Thailand, there are many tools has been used for this task, such as Hamilton Rating Scale for Depression (HAM-D), Mini International Neuropsychiatric Interview (M.I.N.I), Montgomery-Asberg Depression Rating Scale (MADRS), Patient Health Questionnaire-9 (PHQ-9), Thai Hospital Anxiety and Depression Scale (Thai-HADS), Zung Self-rating Depression Scale (SDS), Children Depression Inventory (CDI), Center for Epidemiology Studies Depression Scale (CES-D), EURO-D Scale, Thai Version of the Beck Depression Inventory (BDI), Health-Related Self-Report (HRSR) Scale, Thai Depression Inventory, Depression Screening Test, Thai Edinburgh Postnatal Depression Scale (EPDS), Postpartum Depression Screening Scale (PDSS), and Thai Geriatric Depression Scale (TGDS) [7-9]. Hamilton Rating Scale for Depression (HAM-D) has been commonly used in the study of depression screening and severity evaluation globally. It has high validity and accuracy across various population groups, especially in the study that used an interview methodology [10-11].

AI Technology for Depression Evaluation

Kunugi et al. studied the reliability and validity of the Interactive Voice Response (IVR) program to rate the 17-item Hamilton Rating Scale for Depression (HAM-D) score in Japanese depressive patients. Test-retest reliability of the IVR program was high (intraclass correlation coefficient: 0.93). Internal consistency of each total score obtained by the clinician, psychologists, and IVR program was high (Cronbach's alpha: 0.77, 0.79, 0.78, and 0.83). Regarding concurrent validity, correlation coefficients between total scores obtained by the clinician versus IVR and that by the clinician versus psychologists were high (0.81 and 0.93). The HAM-D total score rated by the clinician was 3 points lower than that of IVR. Inter-rater consistency for each HAM-D item evaluated by the clinician versus IVR was estimated to be fair (Cohen's kappa coefficient: 0.02-0.50) [12].

Gibbons et al. developed a computerized adaptive test (CAT) for depression, called the Computerized Adaptive Test-Depression Inventory (CAT-DI), that decreases patient and clinician burden and increases measurement precision. The 24-item Hamilton Rating Scale for Depression, Patient Health Questionnaire 9, and the Center for Epidemiologic Studies Depression Scale were used to study the convergent validity of the new measure, and the Structured Clinical Interview for DSM-IV was used to obtain diagnostic classifications of minor and major depressive disorder. A total of 1614 individuals with and without minor and major depression were recruited for study. The result showed that the CAT-DI provided excellent discrimination throughout the entire depressive severity continuum (minor and major depression), whereas the traditional scales did so primarily at the extremes (major depression) [13].

Pampouchidou et al. developed an audio and visual depression analysis system. The study used 200 datasets from the Audio/Visual Emotion Challenge of 2013 and 2014 Depression Sub-challenge. By analyzing 68 face keypoints and 70 - 450 Hz audio frequency range compared with the Beck Depression Inventory-II test, the result showed that the system can predict the depression with a high accuracy of 94.8% [14].

Cohn et al. studied the comparison of clinical diagnosis of major depression with automatically measured facial actions and vocal prosody in patients undergoing treatment for depression. Manual Facial Action Coding System (Manual FACS), automated facial image analysis using active appearance modeling (AAM), and pitch extraction were used to measure facial and vocal expression of 57 depression patient interview videos. Accuracy in detecting depression compared with the HAM-D test was 88% for manual FACS, 79% for AAM, and 79% for voice prosody [4].

Oh et al. studied the effectiveness of using the Deep-learning algorithms to assess the factors leading up to prevalence and clinical manifestations of depression. Customized deep-neural-network and machine-learning classifiers were assessed using survey data from 19,725 participants from the NHANES database (from 1999 through 2014) and 4949 from the South Korea NHANES (K-NHANES) database in 2014. The result showed that the deep-learning algorithm trained with serial datasets (NHANES, from 1999 to 2012) predicted the prevalence of depression in the following two years of data (NHANES, 2013 and 2014) with the receiver operating characteristic curve (AUC) of 0.92. In conclusion, the deep neural-networks managed to identify depression well from other health and demographic factors in both the NHANES and K-NHANES datasets [15].

Objectives

1. To develop the first online depression testing service in Thai language that is accessible for the visual disability, and has very high accuracy across all demographic users.

2. To raise awareness of depressive disorder screening and prevention in Thailand, which will help reduce the country’s suicidal rate and increase the economic productivity.