|Topic:||Deep learning for automatic detection of multiple myeloma in CT of femurs|
|Supervisor:||Jan Hering Dipl.-Math.|
|Description:||The diagnostics of multiple myeloma (a type of bone marrow cancer) uses blood plasma tests as well as CT images of large bones (femurs in this case). From the methodical point of view, these data represent a situation to be found more often in next years -- we don't have fully annotated images at voxel level, since this work is very time consuming, but are provided only with a 'global' label (healthy / diseased), which comes naturally from the diagnosis. By means of Multiple-Instance Learning (MIL) we are able to reach reasonable precision in the classification.
However, the interesting question how a deep-learning system will perform is still without answer. The aim of this work is to implement a selected CNN architecture, or to compare multiple architectures, and evaluate it's performance against the current system.
This work thus brings the opportunity to learn more about the newest methods in machine learning, to improve own programming skills as well as to get in touch with the field of medical imaging and computer assisted diagnosis.
|Bibliography:|| J. Hering, J. Kybic, and L. Lambert, “Detecting multiple myeloma via generalized multiple-instance learning,” SPIE Medical Imaging 2018, p. 22.
 F. Martínez-Martínez, J. Kybic, L. Lambert, and Z. Mecková, “Fully Automated Classification of Bone Marrow Infiltration in Low-Dose CT of Patients with Multiple Myeloma Based on Probabilistic Density Model and Supervised Learning,” Comput. Biol. Med., vol. 71, pp. 57–66, Apr. 2016.
|Realization form:||implementation in python|