Histopathology Image Classification for Soft Tissue Sarcoma in Limbs using Artificial Neural Networks

Clinical imaging techniques have been widely used in the classification of cancer biopsy specimen histopathology images of limb soft tissue sarcoma (STS). Here, by automatically differentiating cell patterns in malignant and non-malignant tumors, an efficient classifier based on both accuracy and time requirements is significantly improved, which further reduces intra-inter-obserler variations. Color normalization is carried out using a linear transformation into a grayscale image and the region of interest (ROI) of the image is selected by the pathologist. The wavelet transform has been used to extract statistical texture features (S FT) from the grayscale image of this ROI, and neural correlates with extracted features networks were trained. For the purpose test, the features of a new limb STS tissue sample image are extracted and these extracted values are presented to the already trained networks for classification. In this case, the proposed research uses an artificial neural network (ANN), which leads to prominence by improving the classification methods based on accuracy, sensitivity and specificity. Here, two different types of ANN classifiers are discussed with back propagation neural network (BPNN) and radial basis function network (RBFN) classifiers. Furthermore, here the most significant difference between BPNN and RBFN is analyzed using the receiver operating characteristics (ROC) area under the curve. The performance accuracy of these two classification methods reaches 96.36% and 90.91% for RBFN and BPNN, respectively. Based on these accuracy values, RBFN is found to be more efficient than BPNN classifiers. Finally the cancer cell classification accuracy is increased, decision-making time is reduced, and the initial treatment plan for chronic disease of the limbs tumor has been achieved.