Cooperative Game Theory Leads to More Accurate Leukemia Classification
Flow Cytometry isn’t a word most people hear very often, if at all, but it’s how most cancers are diagnosed in patients across the world. In fact, it is extremely important in the process of accurately recognizing which category of cancers an abnormal blood sample belongs in.
In one study out of Tarbiat Modarres University, Iran, published in August of 2011, Atefeh Torkaman and her colleagues researched the use of Cooperative Game theory to improve their success rate of disease classification within leukemia samples. In the study, CD markers (Cluster of Differentiation markers that can be found on a cell with their own CD number) from cells in each sample, obtained by Flow Cytometry, are sorted. CD markers with values <=20 are given a value of 0, and 1 otherwise, and placed into an NXM matrix. Each marker is a player, and the weights and values of each marker are computed using Shapley values. Three computational axioms, along with the NxM matrix of marker values, are used to determine the precise value of each marker and its importance in causing the disease. Using this process, 8 classes of the disease are produced, and the effect each marker has on a specific class of disease is analyzed by its computed weight.
The study resulted in a 93.12% accuracy rate of identifying the type of leukemia in a sample, as opposed to a 90.16% success rate of correct disease classification using other methods. In practice, these results can be used to create more accurate tests to easily determine the type and quantity of medication used to treat a leukemia patient.
Through this cooperative game theory process, Torkaman is creating a large network of CD markers and the types of leukemia they each cause. In addition, she is cutting out edges between CD markers and leukemia types that are not very strongly correlated with one another, if they’re correlated at all. In effect, this network could be visualized as a bipartite graph, with CD markers on the left and categories of leukemia on the right, and edges between the nodes indicating causation or significance of the CD marker to that specific category. It would be even more interesting to evaluate this graph, and to determine which CD marker has edges to the most categories, and then target that for medicinal development.
http://iospress.metapress.com/content/l65j6x4077288825/fulltext.html