Benchmark Dataset for Automatic Glottis Segmentation
Alternate Titles(s): BAGLS dataset
- Description
The Benchmark for Automatic Glottis Segmentation (BAGLS) dataset contains 59,250 high-speed laryngeal videoendoscopy frames and individually annotated segmentation masks obtained from anonymized recordings of 640 consenting subjects across multiple hospitals in Europe and the United States. Participating institutions were Boston University; Louisiana State University; New York University; Sint-Augustinus Hospital, Wilrijk; University of California, Los Angeles; University Hospital Erlangen; and University Hospital of Munich (LMU). Male and female patients were recruited and participants' ages at the time of recording ranged from 14 to 91 years old.
Laryngeal disorders represented in the dataset include:
- Contact granuloma
- Muscle tension dysphonia
- Paresis
- Muscle thyroarythaenoideus atrophy
- Laryngitis
- Vocal insufficiency
- Papilloma
- Edema
- Leucoplacia
- Insufficient glottis closure
- Carcinoma
- Nodules
- Polyp
- Cyst
Additionally the dataset contains recordings from 380 healthy subjects and 90 subjects with unknown laryngeal health status.
- Geographic Coverage
-
CaliforniaCalifornia - Los AngelesGermanyLouisianaMassachusettsSee all...Massachusetts - BostonNew York (State)New York (State) - New York City
Access
- Restrictions
-
Free to All
- Instructions
- Study data can be downloaded via the Zenodo repository.
- Grant Support
-
DO1247/8-1/Deutsche ForschungsgemeinschaftEC409/1-2/Deutsche ForschungsgemeinschaftZF4010105BA8/Bundesministerium für Wirtschaft und Energie
- Other Resources
-
Jupyter notebook
Contains code for training and analyzing the dataset
Glottis Analysis Tools softwareOptional software for video analysis
Pixel-Precise Annotator tool (PiPrA)Optional software for video analysis