Non-Small Cell Lung Cancer Pathology Slides for STAS Detection Model

129 pathological slides from 91 STAS-positive non-small cell lung cancer patients were used to develop an AI detection framework. The model achieved a Jaccard similarity of 0.846 for tumor segmentation and an average precision of 0.784 for STAS detection. Yu Zhang published this dataset on figshare in 2026 under a CC-BY-4.0 license.

Use Cases

• Training deep learning models for tumor region segmentation based on whole-slide images.
• Developing object detection algorithms to identify Spread Through Air Spaces (STAS) patterns.
• Validating AI model performance against pathologist assessments using interclass correlation coefficients.
• Conducting survival analysis to correlate STAS counts with disease-free survival outcomes.

Strengths

• Includes 129 pathological slides from 91 patients, providing a focused sample for model development.
• Model performance metrics are provided, including a segmentation Jaccard of 0.846 and detection AP of 0.784.
• External validation results are reported, with a precision of 0.670 and recall of 0.705.
• AI-derived STAS counts showed substantial agreement with pathologists (ICC = 0.703).

Limitations

• Row count and column-level documentation are unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the specific patient cohort and institution from which the slides were sourced.
• The dataset is small at 3.1 MB, which may limit the scope of model training or generalization.

Provenance

Source

figshare, author Yu Zhang.

Collection Method

Pathological slides analyzed from STAS-positive NSCLC patients; an AI framework was developed for segmentation and detection.

Time Range

The study period is not specified, but the dataset was published in 2026.

Freshness

Last updated 2026-05-22 17:40:05; freshness should be verified.

Geography

Geographic coverage is not specified.