A Novel Diagnosis and Prognosis Model for the RUNX1 Mutation in AML

Acute myeloid leukemia (AML), characterized by its heterogeneity and low survival rate, is one of the most common types of leukemia in adults, with patients carrying the RUNX1 mutation facing particularly poor prognoses. While management of this disease can be difficult, accurate diagnosis and prognosis can help doctors determine the appropriate treatment options to improve patients' survival chances. Existing machine learning-based AML diagnosis models fail to identify specific mutations with high accuracy. Similarly, existing AML prognosis models do not analyze specific mutations. With this in mind, this research aims to create a novel machine learning-based diagnosis and prognosis system for AML RUNX1 patients. Using a dataset of 5002 blood smear images of AML and healthy patients, a deep convolutional neural network-based diagnosis model was created. The diagnosis model showed an accuracy rate of 95% in identifying a patient's disease status (healthy, AML, RUNX1 AML). The prognosis model uses the novel accelerated oblique random survival forest model R package and was trained on a dataset containing the survival and RUNX1 mutation status of 494 AML patients and synthetically generated data from 100 patients. The model was tested on synthetically generated data from 200 patients and showed that the expected death risk for RUNX1 AML patients accelerates from about 2500 days onwards. These results align with the typical prognosis of RUNX1 mutation, substantiating its accuracy. With additional testing, the machine learning models developed in this study demonstrate the potential to assist clinicians with the diagnosis and treatment of RUNX1 AML patients.