2018-08-012024-05-17https://scholars.lib.ntu.edu.tw/handle/123456789/685139Acute leukemia including Acute myeloid leukemia (AML) and Acute Lymphoblastic Leukemia (ALL) is the most common hematologic malignancies and Myelodysplastic syndrome (MDS) is the most common pre-leukemia abnormalities in Taiwan. Due to the high heterogeneity of the disease, the 5-year survival rate of AML and adult ALL achieved no more than 40% despite profound breakthroughs been made in the treatment of acute leukemia. Therefore, it is crucial to have better comprehensive risk assessment tools by assessing all abnormalities known to associated with treatment failure and tailor-made individual treatment base on the abnormality and molecular/gene profile for every patient. In recent years, a growing number of breakthroughs utilizing AI in clinical research have been reported, for instance, a study of skin cancer classification on skin lesion images using deep-convolutional neural network (CNN) technique achieved performance comparable to dermatologists was reported in early 2017 on Nature. Another outstanding research utilizing deep learning on diabetic retinopathy screening on retinal photography images was published on JAMA in Dec 2016. Therefore, we design this study aiming to develop unbiased, systematic diagnostic and prediction tools for MDS, ALL and AML utilizing machine learning technique. We propose to enroll patients diagnosed with MDS, ALL or AML and treated in National Taiwan University Hospital from 1985 to 2017. We plan to retrospectively collect all available historical exam records including molecular profiles, cytogenetics, flow cytometry (FC), pathology report, medication history, hospitalization record, as well as their clinical outcome as training materials for machine learning-based algorithm development. First, we like to develop the algorithm to differentiate normal, MDS, ALL and AML samples with a probability score so that physicians could reduce the time required to evaluate less complex and low-risk patients and spend more time studying those difficult cases. Furthermore, we like to develop prediction models that could predict those who are likely to progress or survive in the future based on NTUH historical record, and their clinical outcome. The results from this study would provide useful perspectives as new diagnosis and prediction tools and help advance treatment for patients with hematological malignancies.Acute leukemiaAMLALLMDSmachine learningartificial intelligence