Objective sensor-based markers for bipolar disorder

The ITP Research Program on Objective sensor-based markers for bipolar disorder monitoring was established in June 2022 in Honor of Anne de Szczypiorski. This program is aimed to support young researchers and enables them to perform research under the supervision of experienced mentors from academia and the medical sector. The research agenda covers mainly hybrid methods of computational intelligence for data streams and time series, soft computing and computational statistics. This program significantly extends previous work of the founders of the ITP Foundation regarding human-consistent hybrid methods to support monitoring of affective disorders with the use of smartphones. ITP’s ambition is to improve the understandability of the delivered approaches that support diagnosis and monitoring by improved summarization and learning.

This research is performed in collaboration with the Insitute for Psychiatry and Neurology in Warsaw and Systems Research Institute, Polish Academy of Sciences who participated in the prospective observational study (agreement no. KB/1094/17) in years 2017-2018.

The outcomes from this research have been recently presented at the Twentieth International Workshop
on Intuitionistic Fuzzy Sets and Generalized Networks by Filip Wichrowski and Katarzyna Kaczmarek-Majer. Filip Wichrowski has been participating in this ITP program since its establishment. Filip is also preparing a master thesis on this topic. Fingers crossed for his defense in January 2023. The problem considered in this work is the construction of a hidden Markov model for the acoustic features of patients suffering from bipolar disorder. We pose the hypothesis that a hidden Markov model (HMM) with a discrete uniform initial state distribution, a transition matrix estimated via the maximum likelihood method, and emission probabilities described by a normal distribution given the state, is an adequate tool for predicting episodes of BD.
The construction of a hidden Markov model and determining a supervised learning approach for the estimation of its parameters is the main contribution of this work. The discussed slides entitled Hidden Markov Models for phase prediction in bipolar disorder are available under the following link.