Deep Learning Based Automated tool for cancer diagnosis from bone histopathology images

Histopathology procedure is conventional in nature. Moreover, manual procedure of Pathologists can handle only limited cases due to prolonged steps. This manual procedure may mislead the doctor if there are bulk cases to investigate due to time constraint and nature of complicated disease like bone cancer. To deal this research in digital histopathology is vital by developing computer assisted tool for diagnosis. Bone architecture complexity is the main reason to remain as gray area for research. Understanding and exploration of different magnitude of bone anatomy will cater the need for constructing research in automation. This research work framed with three major bone cancer namely Ewing sarcoma, Osteosarcoma and Chondrosarcoma. Deep learning method with the concepts of image processing techniques are adopted for bone cancer detection. Deep learning method studies profound features from lower level to higher level without human intervention. Here, database contains total of 1000 images which are resized into $350 \times 350$ without loss of information. In this study dataset of 600 malignant,200 normal and 200 benign images of same size and resolution are used. The Artificial neural networks model(CNN) is a supervised learning used to analyze abstract features in the training set. Deep Learning demands large number of images and its strength was increased as per requirement by augmentation technique. Contribution of this work is classification of images into categories such as malignant, benign and normal and success rate is 92%.