A physician-in-the-loop approach by means of machine learning for the diagnosis of lymphocytosis in the clinical laboratory

Abstract

The complete blood count (CBC) is one of the most commonly performed test due to its clinical utility, versatility and safety. Thus, there are well established consensus rules to review a CBC through the peripheral blood smear review based on abnormal quantitative values for the CBC parameters or qualitative changes detected by analyser flags, and approximately, 20% of the reviewed samples in clinical laboratories correspond to samples with lymphocytosis. Lymphocytosis classification into benign or neoplastic categories plays a pivotal role not only in the laboratory workflow, but also in the patient management and, surprisingly, about 30% of laboratories fail in such classification. Thus, the current work presents a complete pipeline to develop a machine learning model as an objective aid for the lymphocytosis diagnosis, based on physical-chemical properties of the different leucocyte subpopulations through the exploitation of new cell population data parameters provided by current haematological analysers.

Ähnliche Arbeiten