Radiomics Model for Mycetoma Grains Classification from Histopathological Microscopic Images Using Partial Least Squares Discriminant Analysis (PLS-DA)

18 pages, 41 ref. Mycetoma is a chronic granulomatous inflammatory disease that causes severe deformities, disabilities, with many impact on patients and family, particularly in advanced disease stages or when treatment fails. The therapeutic disease strategy heavily relies on the identification of the causative organism and the corresponding classification of the disease as eumycetoma or actinomycetoma. Various diagnostic tools are used for mycetoma differential diagnosis. Histopathology is considered to be an efficient, cost and time-effective tool for mycetoma diagnosis in endemic areas. While histology is currently, the most used diagnostic tool, it requires well-trained pathologists, and that lacks in most rural areas where mycetoma is endemic. In this communication, we present a computational method to effectively differentiate between eumycetoma and actinomycetoma from the grains features in histopathological microscopic images that is based on Radiomics and Partial Least Squares Discrimination Analysis (PLS-DA). In this work, the data were collected from the Mycetoma Research Center of Khartoum, and the proposed approach achieved mycetoma types identification with an accuracy of 91.8% and 0.836 Matthew's Correlation Coefficient (MCC). This computational tool could be of great benefit in rural areas with limited access to specialised clinical centres.