| Summary: | Accurate diagnosis of blood-related disorders in the field of hematology relies on the precise analysis of white blood cells (WBCs), a task that requires a detailed examination of cell attributes associated with it to classify it properly. While traditional methods are accurate and effective, they can be time-consuming and subject to human error depending on the laboratory technician’s skills. On the other hand, automatic methods like flow cytometers are expensive and require highly trained specialists to operate and be constantly maintained by service engineers.
While there are studies into the categorization of WBC using AI, they are often lacking in necessary morphological details to explain their decision process. This project is supported by Sysmex Corporation and carried out in collaboration with an NTU project team and pathologists, who have developed an innovative approach. It involves identifying 11 morphological attributes for each cell and coupled with introducing comprehensive annotations for a Peripheral Blood Cells (PBC) dataset. Additionally, they have developed a deep learning (DL) model to predict these morphological attributes from cell images. The developed model aims to aid medical laboratory staff in making more informed decisions with explainable artificial intelligence (XAI), enabling quicker and more accurate diagnosis of blood-related disorders. However, the current model has some ongoing issues with predicting nucleus shape correctly.
Therefore, this final year project will focus on improving nucleus shape prediction accuracy and experiment with various techniques like image-preprocessing and attention mechanisms. Any improvement in prediction accuracy will be valuable for medical diagnostics, contributing to the development of a fully functional XAI for healthcare application.
|
|---|