Enhanced Mitotic Figure Detection in Glioma Using Super-Resolution Images and High-Frequency Content Maps

This article presents an advanced approach to enhance mitotic figure classification in glioma histopathology using super-resolution images and high-frequency content maps. Gliomas, a common type of primary brain tumor, require an accurate prognosis for effective therapy, often relying on the identification of mitotic figures. In this study, Convolutional Neural Networks (CNNs) are used to classify cells as mitosis or non-mitosis in glioma tissue samples for prognosis, but the low resolution of extracted cell images can obscure key features, leading to misclassification. To address this, we generate super-resolution images from low-resolution ones and introduce high-frequency content maps to highlight crucial cell shape features through pixel-level variations. An ensemble model, integrating classifiers trained on super-resolution images alongside high-frequency content maps, demonstrates a remarkable accuracy of 99.158% and an F1-Score of 99.156% on test split of used data. The approach demonstrates reliability through five-fold cross-validation and provides an insightful solution to a common challenge in digitized classification tasks.