Mitosis detection on histopathological images using statistical detection algorithms

In this work, the utility and accuracy of the statistical detection algorithms for the detection of mitosis on histopathological images have been investigated. In the first stage, the subset images involving mitotic cells from the original images have been created. The occurance based texture filters have been applied to each subset image. Then the training/testing dataset has been created from these subset images. Later, the three statistical detection algorithms have been implemented in this work, namely matched filtering (MF), constrained energy minimization (CEM) and adaptive coherence estimator (ACE). The accuracies over 80% have been obtained for each method and four different evaluation measures have been utilized. The results indicate that the MF is the best algorithm on mitosis detection among the implemented algorithms.