Recognizing Presence of Hematological Disease using Deep Learning

Accurate classification and counting of blood components is crucial in detection of illnesses of an individual. The widely used methods to count blood components are manual counting and hematology analyzer. With advancement in the field of image processing and machine learning, new and better methods are available for counting and classifying blood components. Deep leaning is training the computer with labelled data for classification tasks. Such techniques have shown high performance and accuracy. Most Deep learning models uses neural network architecture. One of the most popular type of deep learning model is Convolutional Neural Network. CNN convolves learned features with input data, and uses 2D convolutional layers, making this architecture well suited to processing 2D data, such as images. CNN's extract the features from the image automatically using numerous hidden layers. Most Deep learning models use transfer learning that is fine-tuning a pre-trained model. RCNN stands for Region based CNN. Unlike CNN which is used for image classification, RCNN is used for object detection. Thus in this paper, we have proposed a method to classify various components of blood : RBCs, WBCs (Monocyte, Lymphocytes, Eosinophils, Neutrophils and Basophils) and find their count from a microscopic blood image using Faster R-CNN model. Thus generating a CBC (Complete Blood Count) report which can be used by medical professionals to diagnose, suggest tests and treatments to their patients.