BD4QoL adopts multiple Artificial Intelligence methods to investigate the prediction of the long-term effects of cancer treatment from “big data” assets on Head and Neck Cancer patients', based on rich longitudinal information unobtrusively collected through a mobile app, a patient empowerment chatbot, quality of life questionnaires, clinical assessments at follow-up visits and patients' health and socioeconomic data.
BD4QoL will provide classification and prediction models that anticipate the worsening of quality of life or the onset of other health conditions and stratify patients into risk subgroups. Deep Learning based models for sequence modeling like Long Short-Term Memory networks and algorithms for time-series analysis like the autoregressive integrated moving average models will be applied to large datasets from previous studies to create the most accurate predictive model. This model will be enriched with the new behavioral and affective features detected by BD4QoL apps.
Human activity recognition techniques based on a Hybrid Activity Recognition System, including the Action Pattern Discovery and the Pattern-Model Matching algorithms developed by partner University of Deusto, that combine unsupervised learning methods and knowledge-based activity models, are used for behaviors reconstruction modeling and to allow identification of significant deviations and related quality of life deterioration.
The BiDi e-coach chatbot application applies dialogue management based on contextual reasoning and reinforcement learning for the automated delivery of counseling strategies and natural language understanding algorithms for the detection of affective traits embodied in the e-coach and in patient dialogue, through sentiment analysis and emotion analysis technologies. IBM Watson will be leveraged to provide the basis for technical partners to research, train and prototype innovative cognitive and affective computing approaches.