To ascertain the efficacy of the DeepDDH system, a deep learning framework, in enhancing diagnostic accuracy and curtailing follow-up intervals for infants undergoing screening for developmental dysplasia of the hip (DDH), the researchers are executing a blinded, randomized controlled trial. This trial juxtaposes AI-only and AI-assisted assessments of DDH against sonographer interpretations across various proficiency levels in the preliminary analysis of ultrasound images.

1. Participating centers and doctors:

   The data in the ultrasound screening sequence database in this part of the study were mainly from Renji Hospital, Shanghai Jiao Tong University School of Medicine and the Sixth People's Hospital, Shanghai Jiao Tong University School of Medicine. Renji Hospital, the Sixth People's Hospita, and the Pediatric Hospital Affiliated to Fudan University, three top-three hospitals in Shanghai, started Graf ultrasound examination earlier, with an average history of more than 10 years. Each unit can perform Graf ultrasound examination maturely, and has good quality control and management. The technology is relatively mature, and they are responsible for providing 6 expert sonographers with more than 5-10 years of DDH ultrasound diagnosis experience, more than 600 DDH ultrasound operations, and 1 pediatric orthopedic expert with 5-10 years of DDH diagnosis experience to participate in the study. However, several other primary or remote medical institutions with late DDH ultrasound screening and insufficient diagnostic experience were mainly responsible for providing 8 primary sonographers (trained residents who performed 20 DDH ultrasound operations) to participate in the study. The other 4 primary sonographers who participated in the study were provided by the Sixth People's Hospital and the Pediatric Hospital Affiliated to Fudan University, so a total of 12 primary sonographers participated in the study. Before the study, the sonographers involved in this study will be evaluated uniformly and quantitatively through online examination papers.

2. Research process:

One week before the start of the randomized diagnostic trial study, which prospectively reevaluated the retrospective data set of the ultrasound screening sequence database, the sonographers registered in the study received uniform training in the latest DDH ultrasound diagnosis in the form of PPT, video, literature study, and offline instruction.

For the included cases in the ultrasound screening sequence database, they would appear in different control groups in a random form, such as the AI model, the Expert sonographer group (composed of 6 ultrasound experts), the primary sonographer group (composed of 6 primary sonographers), and the primary sonographer (composed of 6 primary sonographers) + AI-assisted group. All the image data in the ultrasound screening sequence database were independently divided into measurement points and measurement lines by each group, and the final diagnosis of α Angle, β Angle and classification results were given by the system.

The random occurrence of cases in the ultrasound screening sequence database has certain rules: (1) According to the date of ultrasound examination, the 2152 ultrasound screenings were divided into 848 days. (2) Block randomization was applied to the 848-day study, using a computer-generated sequence of random numbers, with randomization performed in blocks of four or eight. That is, the cases examined for the first time in each day were divided into one group (AI group, advanced ultrasound expert group, primary sonographer group, AI-assisted primary sonographer group), and the cases for follow-up examination were divided according to the first examination group. Randomization was repeated within each group according to block randomization of 6 or 12 block sizes.

When non-first subsequent follow-up cases were randomly presented, the results of the previous examination were included as a reference. In the AI-assisted group, each sonographer was asked to choose whether to modify or confirm the diagnosis according to the measurement points, diagnostic angles and typing results provided by the AI device. However, in the advanced ultrasound expert group, junior sonographer unassisted group, the dedicated research assistant will turn off the AI display function to ensure that no additional information is provided to the sonographer. The consensus of two ultrasound experts with 10 years of experience in DDH ultrasound diagnosis was used as the gold standard. In case of disagreement, a third pediatric expert with 5-10 years of experience in DDH diagnosis and treatment was used to evaluate the diagnosis and treatment of DDH. The final consensus was used as the gold standard.

Finally, the operation results of the above different groups were summarized and analyzed by independent research assistants, including α Angle, β Angle, typing results, and the specific follow-up experience of the case (including follow-up times and follow-up time).

Inclusion Criteria:

• Infants aged 28 days to 6 months who underwent DDH ultrasound examination in Renji Hospital, Shanghai Jiaotong University School of Medicine and the Sixth People's Hospital, Shanghai Jiaotong University School of Medicine between August 2014 and December 2021

Exclusion Criteria:

• Patients lacking or incomplete ultrasound images;
• Patients with poor image quality and unusable images after assessment, including non-compliance with Anatomical identification (Checklist I consists of seven anatomical structures: 1. Chondro-osseous border, 2. Femoral head, 3. Synovial fold, 4. Joint capsule and perichondrium, 5. Labrum, 6. Cartilagineous roof, 7. Bony roof) and Usability check (Checklist II includes three anatomical landmarks: 1. Lower limb of the os ilium, 2. Parallel middle plane, and 3. Labrum);
• Infants with hip dysplasia caused by other diseases such as cerebral palsy, joint contracture, suppurative coxitis, etc., or with other hip diseases and limb deformities.

The clinical demographic information of the subjects was recorded through the Case Record Form, and the data was collected and managed electronically through the ResMan platform.