In recent years, image inspections have become a useful tool in the early diagnosis of rheumatoid arthritis (RA). However, evaluations in RA ultrasound inspections are visual inspections by doctors, evaluation tends to become somewhat subjective, meaning that it is difficult for less-experienced doctors to make an assessment. In this study, therefore, we propose a system that alleviates the burden on doctors, by automating RA image diagnosis, that is to say an RA image diagnosis support system using ultrasound images. With this system, as it is necessary to automate RA image diagnosis, it is necessary to propose an image classification method with higher classification accuracy. Therefore, in this paper, we propose an image classification method with higher precision, by comparing methods of image classification. We also show our evaluation results. As a result, it is considered that image classification via CNN using automatic extraction images of synovial thickening surrounding areas is effective for image classification of ultrasound images for the purpose of automating RA image diagnosis.