An automatic diagnostic network using skew-robust adversarial discriminative domain adaptation to evaluate the severity of depression. (May 2019)

• Record Type: Journal Article
• Authors: Sun, Bo; Zhang, Yinghui; He, Jun; Xiao, Yongkang; Xiao, Rong
• Is Part Of: Computer methods and programs in biomedicine. Volume 173(2019)
• Page Start: 185
• Page End: 195
• Publication Date: 2019-05
• Subjects: Deep learning -- Domain adaptation -- Skew-robustness -- Adversarial learning -- Knowledge evolution; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2019.01.006
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: We propose a novel automatic depression evaluation approach, called SRADDA, that can minimize domain discrepancy distance using a skew-robust adversarial objective in a domain discriminator when shifting from a large-scale source domain, the Twitter sentiment analysis training corpus (TSATC), to a small-scale target domain, the distress analysis interview corpuswizard of oz(DAIC-WOZ) database. We build a top-down selection mechanism that can automatically extract key answers concerning about certain topics, e.g. sleep status, PTSD/depression diagnosis, successive treatment, personal preference, and feelings from interview documents. To solve the problem of differing label spaces across different datasets, e.g. the Twitter and depression interview text sets, we construct a feature projector that can project the key answer embeddings to further discriminate and interpret semantic binary features. Abstract: Background and Objective: Deep learning provides an automatic and robust solution to depression severity evaluation. However, despite it is powerful, there is a trade-off between robust performance and the cost of manual annotation. Methods: Motivated by knowledge evolution and domain adaptation, we propose a deep evaluation network using skew-robust adversarial discriminative domain adaptation (SRADDA), which adaptively shifts its domain from a large-scale Twitter dataset to a small-scale depression interview dataset for evaluating the severity of depression. Results: Without top-down selection, SRADDA-based severity evaluation network achieves regression errors of 6.38 (Root Mean Square Error, RMSE) and 4.93 (Mean Absolute Error, MAE), which outperforms baselines provided by the Audio/Visual Emotion Challenge and Workshop(AVEC 2017). However, with top-down selection, the network achieves comparable results (RMSE = 5.13, MAE = 4.08). Conclusions: Results show that SRADDA not only represents features robustly, but also performs comparably to state-of-the-art results on small-scale dataset, DAIC - WOZ .