Assessment of small bowel’s motility plays an important role in the diagnosis of small bowel disease. Conventional assessment methods rely on hand-designed features or manual guide-line drawing, which results in difficult modeling and high time consumption, thus they are still inefficient and impractical for clinical uses. With the help of deep neural networks, we introduced a semi-automated approach, replacing hand-designed features with automatic feature extraction, and an automated approach, eliminating manual guide-line drawing, to assess small bowel motility by automatically marking cross-sectional diameters on small bowel images, measuring temporal fluctuation of diameter lengths, and evaluating contraction frequency. Experiment results show that proposed methods could estimate small bowel contraction frequency correctly. The difference between predicted diameter lengths and one manually labeled is within reasonable range, and estimated frequency is close to the groundtruth. Therefore, proposed methods can be utilized for diagnosis of small bowel disease, which will assist radiologist in decision-making. [ABSTRACT FROM AUTHOR]