In recent years, blood cell classification with the help of image processing techniques has attracted many researchers to build an automated system that assists doctors for diagnosis of cancer. Also, it’s very challenging to differentiate cancer cell from normal cell as they look similar in initial stages. In this manuscript, we have presented an approach for cancer cell detection by extracting important features from the blood cell images and learning multiple classifiers. We have observed that Gradient Boosting Decision Tree classification algorithms give better result than Support Vector Machine. We have also derived few important features like presence of adjacent nuclei and measure of irregularity in the shape of a nucleus, which has significant impact on cancer cell detection. Our techniques can be used in a limited computing environment without a Graphics Processing Unit.
We have achieved 85.6% of F1 score on validation data. This approach also identified an important feature for the images that can help doctors or technicians for better understanding of stained images to aid diagnosis of leukemia patients