MosMedData: Chest CT Scans With COVID-19 Related Findings Dataset

TL;DR

AI development requires large, representative clinical datasets, yet available COVID-19 CT datasets are often small and lack region-of-interest annotations. This dataset provides anonymised Moscow hospital lung CT scans spanning COVID-19-related findings and no such findings, with a small expert-annotated subset, for education, calibration, and independent algorithm assessment.

• Problem

  Available datasets are relatively small and rarely include additional information such as tags or binary masks for regions of interest, despite the need for large, representative clinical data.

• Method

  The dataset collects anonymised human lung CT scans from municipal hospitals in Moscow, including CT0–CT4 findings, and expert annotations of ground-glass opacifications and consolidations for 50 studies.

• Results

  The dataset is intended for education, calibration, and independent assessment of AI computer-vision algorithms.

• Takeaways & Limitations

  The dataset supports AI applications including patient triage, worklist prioritization, rapid assessment of abnormal changes, and reducing missed abnormalities.

• Takeaways & Limitations

  Only a small subset of studies, comprising 50 cases, has expert annotations with binary masks for the regions of interest.