The analysis of blood tests is usually done by using automated instruments that determine the main characteristics (haemoglobin level, number of red blood cells, white blood cells) of the sample. In 10% of cases however, an additional examination is required: in this case, an expert has to analyse scans of the slides to classify different elements of the blood and detect any abnormal situation. This approach can be automated using Artificial Intelligence methods. In the past, the automatic classification of abnormal white blood samples has been performed, but the process needs to be improved by including all blood elements in the analysis, by speeding up the method, and by improving the possibility of generalising results for different acquisition conditions (artefacts, colour, light, scanner type).
The objectives of the PhD will be to first create a substantial dataset of annotated images, then to automatically detect different elements of the blood with potential abnormalities, and finally to classify nucleated blood elements (globules), for example into different categories of pathological white blood cells. At the end of the project, the solution must be designed to be fast, and must be able to generalise for different image acquisition conditions. In addition, special attention will be paid to improving the learning phase, handling the large number of classes, and deciding how the algorithm will perform several tasks in parallel. Finally, the diagnosis will be adapted to a patient level, considering the sample as a whole, and not as a categorisation of individual elements.
PhD student: Manon CHOSSEGRO
PhD supervisors: Xavier Tannier, Daniel Stockholm
Research laboratory: LIMICS - Laboratoire d'Informatique Médicale et d'Ingénieurie des Connaissances en e-Santé