Computer-aided therapeutic diagnosis for anorexia

Abstract

Background: Anorexia nervosa is a clinical disorder syndrome of the wide spectrum without a fully recognized etiology. The necessary issue in the clinical diagnostic process is to detect the causes of this disease (e.g., my body image, food, family, peers), which the therapist gradually comes to by verifying assumptions using proper methods and tools for diagnostic process. When a person is diagnosed with anorexia, a clinician (a doctor, a therapist or a psychologist) proposes a therapeutic diagnosis and considers the kind of treatment that should be applied. This process is also continued during therapeutic diagnosis. In both cases, it is recommended to apply computeraided tools designed for testing and confirming the assumptions made by a psychologist. The paper aims to present the computer-aided therapeutic diagnosis method for anorexia. The proposed method consists of 4 stages: free statements of a patient about his/her body image, the general sentiment analysis of statement based on Recurrent Neural Network, assessment of the intensity of five basic emotions: happiness, anger, sadness, fear and disgust (using the Nencki Affective Word List and conversion of words to their basic form), and the assessment of particular areas of difficulties-the sentiment analysis based on the dictionary approach was applied. Results: The sentiment analysis of a document achieved 72% and 51% of effectiveness, respectively, for RNN and dictionary-based methods. The intensity of sadness (emotion) occurring within the dictionary method is differentiated between control and research group at the level of 10%. Conclusion: The quick access to the sentiment analysis of a statement on the image of patient's body, emotions experienced by the patient and particular areas of difficulties of people prone to the anorexia nervosa disorders, may help to establish the diagnosis in a very short time and start an immediate therapy. The proposed automatic method helps to avoid patient's aversions towards the therapy, which may include avoiding patient-therapist communication, talking about less essential topics, coming late for the sessions. These circumstances can guarantee promising prognosis for recovering.

Investigating machine learning and natural language processing techniques applied for detecting eating disorders: a systematic literature review

“…Moreover, Bag of Words (BoW) and various types of Word Embeddings, including GloVe ( 35 , 48 ), FastText ( 35 ), and Word2Vec ( 35 , 36 ), were widely employed as feature extraction techniques in these studies.…”

Determining the Intensity of Basic Emotions among People Suffering from Anorexia Nervosa Based on Free Statements about Their Body

“…The authors of papers [21][22][23][24] used systems in their research based on various classifiers and methods for feature extraction. The recurrent neural network (RNN) classifier is often used to classify basic emotions.…”

An analysis of patient motivation for seeking online treatment for binge eating disorder—A mixed methods study combining systematic text condensation with sentiment analysis

“…Despite this general definition, however, most research and algorithms developed in the field of natural language processing focus on polarity detection, which aims to represent the emotional content of linguistic entities as scores on a scale from strongly negative to strongly positive (40)(41)(42)(43)(44)(45)(46)(47)(48)(49)(50). Now a fundamental tool of computational text analysis, polarity detection (henceforth polarity) has notably been used in psychology for analyzing sentiment in dreams (51), detecting mental disorders in Tweets (52), diagnosing anorexia nervosa (53,54), and assessing the wellbeing of Tweeters during the pandemic (55).…”