Loading...Loading...

Using Artificial Intelligence to identify image biomarkers of mature White Blood Cells from peripheral blood smears.

Carvalho Gonçalves, José

(2021)

Details

Supervisors

John Lee

Faculty

Faculté de médecine et médecine dentaire

Degree label

Master [180] en médecine

Abstract

This thesis introduces a novel workflow whose aim is to identify biomarkers from medical images. The proposed approach is made up of two-steps. In the first step, we automate the cell analysis and in the second step, we try to identify biomarkers among several radiomics (or textural) features and more traditional morphologic (or geometric) features. Even though such approach can be conceptually applied to a large variety of problems, we chose mature White blood cells (WBC) from peripheral blood smears, mainly because the issue has been largely neglected by the AI community (see section 2.2) despite the clinical and educational relevance (see section 6). Despite the analysis of more than 100 features and almost 55.000 images, the proposed workflow did not identify new image biomarkers of WBC. Nevertheless, it allowed us to verify commonly used cell characteristics values and establish new gold-standard norms for WBC. These results can, however, be criticized because of the data selection bias. Yet, the major issue of our work is the feature selection process (step 2) and hence, one should improve it before implementing our workflow to more ambitious tasks such as the diagnosis of malaria and hematological diseases – of particular usefulness for undeveloped countries – and for educational purposes by, for instance, helping biomedical students during their histology courses.