Collaborative Web-Based Platform for Clinical ATS Triage Labeling

Abstract

In low-resource environments, developing high-quality labeled datasets for clinical triage is a challenging task due to the limited availability of experts and the difficulties associated with managing collaborative annotation when multiple labelers are working at the same time. To deal with these problems, a web-based annotation platform was developed that hosted the MIETIC Indonesian clinical triage corpus. This platform was built on a serverless architecture using Streamlit and Google Sheets, which includes a hybrid concurrencycontrol mechanism. This mechanism uses process-level threading locks in combination with an optimistic cell-level verification protocol to provide data consistency without the need for a dedicated transactional database. The evaluation of the platform included two workloads using 5 and 10 labelers, respectively. The results show good scalability, with the number of completed records increasing from 492 to 753 while the cellwrite success rate remained at or above 97.5% and no API retries were recorded. The platform also includes automated data management and quality-control workflows that streamline the annotation process. By bridging the gap between ad hoc manual practices and more complex enterprise annotation systems, the proposed solution offers a practical, scalable, and reliable approach for collaborative clinical NLP dataset development in resource-constrained environments.