Leukemia Detection with Overlapping Blood Cells Using Watershed Algorithm and Convolutional Neural Networks

Various techniques in image processing are generally utilized in the field of medicine, considering the factor of health, therefore it must be contrived in a circumspect manner. The study was developed to analyze and identify if the Watershed Algorithm will affect the accuracy in classifying blood smear images as a candidate for Leukemia or not. The proponents constructed a system using a Convolutional Neural Network to classify the input blood images. The system will process both batches of raw input blood images and with separation of overlapping blood cells, through image processing techniques and Watershed Algorithm, before proceeding to the image classification algorithm. The system was trained and tested three times in both cases to ensure the reliability of the Convolutional Neural Network classification model with and without the Watershed Algorithm applied. The system without the Watershed Algorithm applied yielded an accuracy of 90%, 88%, and 90%, respectively with an average of 89.33% while the system with Watershed Algorithm applied yielded an accuracy of 96%, 94%, and 94%, respectively with an average of 94.67%.