Thesis title:
Speech and oculomotor features of Cancer-Related Cognitive Impairment
Abstract:
This dissertation sheds light on Cancer-Related Cognitive Impairment (CRCI). This condition refers to a set of cognitive changes reported by cancer patients (excluding cancers of the central nervous system). One of its predominant symptoms is word-finding difficulties. Although it can affect up to 75% of cancer patients, CRCI remains challenging to diagnose using current neuropsychological tools. This dissertation aims to identify speech markers that can detect CRCI, and to better understand the cognitive mechanisms underlying this disorder. Forty-four breast cancer survivors and 13 healthy control participants participated in this study. The participants were instructed to tell a picture-based story. Speech productions were recorded and transcribed using semi-automated methods. Speech breaks (i.e., silent pauses, filled pauses, and sustained vowels) were annotated. Fourteen speech parameters were preselected and incorporated into machine learning models. In addition, the oculomotor behavior of 8 survivors and 13 controls was recorded using an eye-tracking technique. Heatmaps were generated based on fixation point coordinates. Finally, all participants underwent a psychological and cognitive evaluation. The relationship between speech data and other variables was examined using Bayesian statistics. Machine learning models identified a subset of three speech markers of CRCI: speech-to-silence ratio, silent pause mean duration, and filled pause mean duration. The combination of these three parameters enabled the classification of breast cancer survivors with CRCI, survivors without cognitive complaints, and controls with an accuracy of 73.7%. Analysis of the eye-tracking data showed a significant delay between the first gaze at an object and its naming in cancer survivors. This result suggests that survivors struggle to access their lexical stock. Moreover, the fixation maps suggest that cognitive deficits in CRCI may be more associated with a deficit in executive control than in memory problems. However, heatmap analyses do not rule out possible attentional deficits affecting lexical search. This dissertation offers new insights into speech-based diagnostic tools. This research will be extended through further studies on other cancer types and conditions with subtle cognitive impairment.
Thesis supervisors:
Pr. Fabrice HIRSCH, PhD (PU, Université Paul-Valéry Montpellier 3, Praxiling UMR 5267 ) & Pr. Sophie JACQUIN-COURTOIS, PhD, M.D. (PU-PH, Université Lyon 1, Centre de Recherche en Neurosciences de Lyon UMR 5292 INSERM U1028 ; Hospices Civils de Lyon)