Multi Instance Learning : Classification of Lymphocytosis from Blood Cells
Lymphocytosis is a common finding, which can be either a reaction to infection, acute stress, and so on (termed reactive), or the manifestation of a lymphoproliferative disorder --a type of cancer of the lymphocytes (termed tumoral). In existing clinical practice, diagnosis as either reactive or tumoral relies on visual microscopic examination of the blood cells together with the integration of clinical attributes such as age and lymphocyte count. Taking into consideration the visual assessment based on clinical attributes together with texture and size of the lymphocytes in the blood smear, a diagnosis of the subtype of lymphoid malignancy is performed. On the positive side, such practice is fast and affordable. It suffers however from poor reproducibility. Additional clinical tests are required, with flow cytometry being the gold standard to definitively affirm the malignant nature of the lymphocytes. However, this analysis is relatively expensive and time-consuming, and therefore cannot be performed for every patient in practice. Therefore, the development of automatic and accurate processes could lead toa better way to determine which patient should be referred for flow cytometry analysis, augmenting and assisting the assessment of the clinicians.
To build a dataset for this problem, blood smears and patient attributes were collected from 204 patients from the routine hematology laboratory of the Lyon SudUniversity Hospital. The samples were anonymized as required by the General Data Protection Regulation, keeping basic demographic information, age and sex, intact. The inclusion criteria were (a) a lymphocyte count above 4×10^9/L, and (b) absence of opposition to the research. The blood smears were automatically produced by a Sysmex automat tool, and the nucleated cells were automatically photographed with a DM-96 device. In particular, you will have access to 142 subjects with 44 reactive and 98 malignant cases for training and 42 subjects for testing.
In this folder, you will find some notebooks to deal with multi instance learning. More details can be found in the folder's readme file.