Significant Feature Extraction Automated Framework for Cancer Diagnosis from Bone Histopathology Images

Osteosarcoma and Ewing Sarcoma most widely recognized primary bone tumor today. Due to tissue structure complexity there is limited research to explore for digital automation in histopathology field. Different parts of tissues in bone biopsy are depicted with osteoblasts, osteocytes, and osteoclasts. In these cells common malignant features are nuclear membrane irregularities, pleomorphism, multinucleated giant cells, hyperchromatic nuclei and abnormal mitoses. Aforesaid feature extraction to improve diagnosis accuracy is the prime objective of this study. This paper presents a novel image analysis frame work for ascertain malignancy level in one of the three predefined class. A total of 200 Hematoxylin and Eosin (H&E) stained histopathological images comprising of osteosarcoma, Ewing Sarcoma and normal bone (includes benign) are used in this study. The proposed work has following elements. First, the given datasets are classified into three groups based on different features. Secondly, automation framework which executes a new object and color based segmentation method to recognize all tissue cells. Moreover, it extracts all the important features such as malignant osteoid, hyperchromatic nuclei and nuclei count from the segmented images and partitioned into three objective groups. Next, training set with low level descriptors are derived for each group and used to instruct a Support Vector Machine (SVM). Finally, the approach derives test sets with unknown class label are given to trained SVM classifier. Depending on the result of binary classifiers the algorithm accurately determines the malignancy level. This experiment shows high quality result with classification accuracy 93.7%.