Hjem
Klinisk institutt 1
Midtveisevaluering

Midtveisevaluering - Petter Jakobsen

Midtveisevaluering for ph.d.-graden ved Universitetet i Bergen for kandidat Petter Jakobsen

Petter Jakobsen er tilknyttet Klinisk institutt 1. Veiledere er Ketil Joachim Ødegaard, Ole Bernt Fasmer, Tine Nordgreen og Jim Tørresen.

Prosjekt

Detecting and predicting mood transitions in bipolar disorder using motor activity

Abstract

Background
Disrupted biological rhythms are characteristic symptoms of bipolar disorder, and recordings of motor activity is an objective method for observing the biological state. Although motor activity is far from substantially studied, bipolar patients are found to differ from healthy controls in measures of variability, predictability, and complexity in activity patterns. In biological systems, early warning signs occur before marked changes in the state of the system, so called critical transitions. Evidence support the existence of a critical transition preceding change of mood state in bipolar disorder.

Method
The first paper of the PhD-thesis will be a reanalysis previous presented data of depressed patients compared to healthy controls. The second and third paper will use data from an ongoing study. Hospitalized bipolar patients with agitated features of a manic episode will have motor activity recorded twice for 24 hours, first at inclusion to the study and secondly at discharge from the hospital (when in remission). Then, together with additional included bipolar patients followed up for a year with continually recording of motor activity.  Mathematical tools obtained from the field of non-linear systems, complexity theory and chaos theory are required to analyze and evaluate motor activity time series. Various machine-learning techniques have also displayed promising results in analyzing complex dynamical systems data.

Results
In the first paper (soon to be submitted) both Random Forest and Deep Neural Network machine learning techniques presented promising abilities in discriminating between depressed patients and healthy controls in motor activity time series.

In preliminary complexity analysis of the dataset for the second paper, motor activity shows intra-individually differences for patients with bipolar disorder in manic and euthymic phases

Paper 3 will utilize data from the one year of continues motor activity recordings to identify critical transitions preceding onset of mood episodes.

Conclusion
Various machine learning techniques presents promising abilities in discriminating between depressed patients and healthy controls in motor activity. The bipolar manic state seems intra-individually associated with augmented complexity, reduced variability and deviating circadian rhythms compared to euthymia in motor activity. Preliminary results needs validation through paper 2 (and paper 3) before final concluding is possible.