Immunohistochemical analysis of oral cancer tissue images using support vector machine

This paper describes an automatic image analysis technique for p53 immunostained tissue sections of oral cancer. The tissue images are segmented using the entropy thresholding and clustered cells are resolved by selectively applying watershed transform. Each cell nuclei of tissue images is classified as positive or negative according to the staining intensity using support vector machine, and then, tissue score is determined as per J-scoring protocol. The performance of the feature and also scoring technique has been evaluated separately by an individual dataset. According to the experimental result, the feature extracted from the blue component has attained the highest classification accuracy of 98.01% with sensitivity and specificity of 98.86% & 94.74% respectively. The outcome of automatic technique based on the blue component has a strong agreement with the manual score. Therefore, automatic tissue scoring has high potential in the field of modern cancer diagnosis and specific therapy design for the patients. •An automatic tissue image analysis technique has been presented.•Nuclei segmentation is performed by entropy thresholding and watershed transform.•Maximal separation feature is extracted from nuclei for classification.•Automatic J-score is determined and compared with manual score.•The outcome based on the blue component has a strong agreement with the manual score.