Deploying Unsupervised Learning for Daily Activity Windows Analysis in Cancer Patients

Radiotherapy (RT) could have significant impact on quality of life (QoL) in cancer patient. The aim of the study is to evaluate Radiation Treatment impact on daily activities in cancer patient using wearable devices technologies to monitor patients in their daily activities. Objective data were collected over RT time and a dedicated in-house algorithm was designed to recognize and analyze several repeated activity windows (RAWs) during days based on the intensity of the subject's activity (steps walked) and the duration of such activity (minutes). Subjective patient reported outcomes (PROs) were collected in the form of an answer scoring to a dedicated questionnaire. Unsupervised machine learning approach (K-means) was used to cluster RAWs over RT time for each patient. 9–18 RAWs were counted in a day per patient, by the conducted analysis. Four RAWs clusters were found characterized by a variation in their parameters. The analysis of the clusters' parameters showed that subjects' activity level could be corelated with the variation in the subjective PRO providing indicators of the RT treatment impact on the patients' QoL. QoL variation evaluation could be improved by blending the objective data collected via a continuous patient monitoring with the powerful information coming from the PRO.