Soft tissue tumors arise in soft tissues in the body. Because the soft tissue tumor is one of the rare cancers, it is difficult to diagnose and treat in general hospitals. This study focuses on obtaining a quantitative index that physicians can use to build a treatment plan for soft tissue tumors based on the patient’s survival period or risk of experiencing clinical events in the future. This paper proposes a convolutional neural network-based patient’s survival period prediction method using pathological images. Since soft tissue tumors are rare cancers, the number of subjects is not enough to train a complex CNN model. We used the mean-variance loss to tackle the regression task predicting the survival period using ResNet18, a classification CNN model. Furthermore, we improved the soft label to efficiently train the classification CNN for the regression task. Experiments on 44 whole slide images of 44 patients showed that the improved soft label called fit label achieved the lowest mean absolute error of 6.5 months and the highest concordance index of 0.893. From these results, the proposed method can be used to evaluate the risk level of a patient with soft tissue tumors.