Sputum M-ROSE Images for Bacterial Detection via YOLOv6, 161 Samples

161 Gram-stained sputum M-ROSE images containing Acinetobacter baumannii, Klebsiella pneumoniae, and Pseudomonas aeruginosa were collected using a digital scanner. The dataset was created by Yan Huang and last updated in April 2026 to support a study on few-shot and transfer learning for bacterial identification. It includes 30 annotated images for training and validation, with 131 images reserved for independent testing.

Use Cases

• Training few-shot learning models for bacterial identification based on the limited annotated images mentioned.
• Evaluating transfer learning performance of object detection architectures like YOLOv6 based on the described framework.
• Benchmarking custom data augmentation strategies like rotation cutting and image fusion for small target detection.
• Developing automated tools for rapid on-site microbiological evaluation (M-ROSE) to reduce clinician interpretation time.

Strengths

• Includes 161 real-world clinical Gram-stained sputum images.
• Model trained on the data achieved 93.89% accuracy on an independent test set of 131 images.
• Dataset is openly shared under a CC-BY-4.0 license.
• Research methodology and performance metrics (recall, precision, F1, mAP) are explicitly documented.

Limitations

• Dataset is small in scale at 661.4 KB, limiting its use for training large models from scratch.
• Row count and column-level documentation are unknown; data structure must be inspected after download.
• Focus is limited to three bacterial species (A. baumannii, K. pneumoniae, P. aeruginosa).

Provenance

Source

figshare, author Yan Huang.

Collection Method

Images collected using a digital scanner from sputum M-ROSE smears.

Time Range

null

Freshness

Last updated 2026-04-22 05:46:36; freshness should be verified.

Geography

null