Lymphocytosis classification
Data challenge on medical image classification. Project done during my master’s degree.
The goal of the challenge was to predict the reactive or malignant nature of lymphocytosis in patients. Each patient is represented by a set of images of lymphocytes of arbitrary size, and by patient-level features (age, lymphocyte count, etc.). This project explores two main approaches to the problem:
An unsupervised classification approach where we cluster the images of lymphocytes and use the cluster assignments as features for a classifier.
A mixture of experts approach where we separetely classify the images and the patient-level features and learn a mixture mechanism to combine the two predictions.
Wouter Van Gansbeke, Simon Vandenhende, Stamatios Georgoulis, Marc Proesmans, and Luc Van Gool. SCAN: Learning to Classify Images without Labels. arXiv (Cornell University), 5 2020. URL http://export.arxiv.org/pdf/2005.12320.
Mihir Sahasrabudhe, Pierre Sujobert, Evangelia I. Zacharaki, Eugenie Maurin, Béatrice Grange, Laurent Jallades, Nikos Paragios, and Maria Vakalopoulou. Deep Multi-Instance learning using Multi-Modal data for diagnosis of lymphocytosis. IEEE Journal of Biomedical and Health Informatics, 25(6):2125–2136, 6 2021. doi: 10.1109/jbhi.2020.3038889. URL https://doi.org/10.1109/jbhi.2020.3038889.
- The code is available on github.
- The project report is also available on github.