<p>Medical image intelligent analysis systems using deep learning models for lesion segmentation in medical images, such as X-ray, CT, MRI, and ultrasound could assist doctors to improve the efficiency of clinic diagnosis and treatment, especially in the early diagnosis of diseases such as pulmonary fibrosis. The challenge for quick response of lesion segmentation under environment of a large number of patient visits is that AI models need to handle a large number of inference computation, for instance, a set of lung CT scan of a patient usually contains more than 300 images. This paper proposes a service orchestration method, aiming to address the above challenge by improving the parallelization of the process of AI model inference at a smaller granularity which is called as operator. The proposed operator-based method includes phases of AI model operarization, operator instance composition, and operator dynamic scheduling. Firstly, complex deep learning models are decomposed into independent operators. Then, operator instances are combined to form the whole inference process of an AI model in a Directed Acyclic Graph (DAG) structure. Finally, considering both the logical dependency implied in the DAG and the computational resources required for operator execution, a scheduling algorithm is designed to automatically parallelize the whole process of lesion segmentation. The feasibility of the method is validated through a case study of a pulmonary fibrosis diagnostic support system, demonstrating its potential in improving operational efficiency of medical image intelligent analysis systems.</p>