How an AI Code Assistant Cut Medical Imaging Data Processing Time by 9×

Graduate student Wang Rongsheng and his research lab were overwhelmed by repetitive coding tasks required to process 150 GB of medical imaging data (DICOM to PNG conversion, window slicing, and JSON metadata generation), which previously needed three people working for a full week.

After learning about Baidu Comate—a large‑model‑based AI coding assistant integrated into VS Code, JetBrains, and Xcode—they adopted its features such as real‑time code continuation, natural‑language‑driven code generation, comment‑to‑code, and private‑knowledge enhancement.

Using a single natural‑language prompt like

"Read all medical images in the image folder, split each 512×512 window with overlap, record window info in a JSON file for reconstruction, and apply the same process to label images, saving results to image_output and label_output folders."

, Comate generated the full Python pipeline, cutting the development time from one week to two days for one person.

The tool also streamlined other workflow steps: it quickly generated code to convert annotation masks into JSON format upon client request, automatically added standardized inline comments, and facilitated team collaboration by ensuring consistent documentation across contributors.

With the newly released Comate Open Platform, the team connected their personal “code repository” to the assistant, enabling private‑knowledge enhancement that improved code generation accuracy by 10‑20 % and further reduced manual searching.

Overall, Baidu Comate enabled a more than nine‑fold increase in processing efficiency, allowing the lab to focus on higher‑level research while the AI handled routine coding tasks.