Survival Time Estimation for Soft Tissue Tumors Using Pathological Images with Label Distribution Learning

Abstract

Number of patients of the soft tissue tumor is relatively very small to the number of well-known cancers. The pathological diagnosis is essential for the treatment of malignant tumors, but there are few pathologists familiar with malignant soft tissue tumors or medical institutions specializing in their diagnosis. Pathological images are used for postoperative treatment planning and to determine whether chemical treatment should be administered. However, this is too difficult to decide for physician, so it is burden on. Therefore, an objective diagnostic system based on pathological images is required. This paper proposes survival time estimation methods using classification and label distribution learning for soft tissue tumor patients. We believe that this will realize treatment tailored to the patient's medical condition. In this paper, we propose a method that uses the label distribution learning in survival estimation can improve accuracy. The minimum MAE for soft label was 7.69 months and compared to hard labels, the maximum improvement was about one month.