A trichotomie technique to separate overlapped nuclei in microscopic cancer images

Ability to clearly delineate the nuclei of microscopic cancer cells is crucial to the accuracy and efficiency of image-based approaches to cancer diagnosis and treatment. Oftentimes, however, such cells contain overlapped (or touched) nuclei. The study proposed in this work presents a hybrid trichotomic technique that combines the Gram-Schmidt method (GSM), handling of relevant geometric features of the cell nuclei, and application of the K-means clustering algorithm to segment, detect, and separate touched nuclei in microscopic cancer images. Using a dataset of microscopic images from two datasets comprising of breast cancer cells and acute lymphoblastic leukemia the proposed technique achieves average mean square error (MSE) of 0.087 and 0.075 for the two datatypes, respectively. Utilising the K-means clustering algorithm in the separation phase of the proposed technique ensures an average normalized accuracy of 0.73 and 0.91 respectively in terms of the nuclei separation for the microscopic breast cancer and acute lymphocyte leukemia cell images in comparison to manual approaches.