Search
Loading

Automated classification of Acute Myeloid Leukemia

Computational models of acute myeloid leukemia immunophenotypes

Acute myeloid leukemia (AML) is a malignant disease with an age-adjusted incidence of 3.51 per 100,000 men and women per year in the United States. The 5-year relative survival is only 25% for patients diagnosed between 2004 and 2010. For elderly patients (65+), the 5-year survival rate is less than 6.0%. Cytometry immunoptyping is the leading technology for cell analysis, allowing rapid evaluation of heterogeneous cellular populations in a single-cell setting, i.e., interrogating separately every individual cell in a sample. My research uses a unique computer-aided diagnostics (CAD) methodology employing a modern nonparametric Bayesian approach and incorporating the latest developments in machine learning and statistical model building to automate the analysis of clinical immunophenotyping data in the context of AML. The major obstacle to the robust implementation of the machine-learning approaches for the classification of clinical data arises from the complexity, variability, and diversity of aberrant immunophenotypes available for model training. Consequently, any learning with an abundance of normal cases and a limited set of specific anomalous cases becomes biased toward the specific types of immunophenotypic anomalies represented in the training set. Such models suffer from overtraining and do not accurately identify samples representing disorders. Although one-class classifiers trained using only healthy cases would avoid such a bias, robust sample characterization is still critical for a generalizable model, which can become the basis for CAD. Yet, owing to sample heterogeneity and instrumental variability, arbitrary characterization of samples by feature extraction algorithms usually introduces feature noise that typically leads to poor predictive performance. To solve this fundamental problem, I use an original nonparametric Bayesian approach that explicitly assumes the presence of random effects (owing to biological, instrumental, and technical variability) and identifies phenotypic differences across batches of samples in the presence of these effects

Page updated on 2022-12-07 23:32:00 -0500