CystoDS: a multiclass endoscopy image dataset for artificial intelligence-assisted bladder cancer detection.

Area of Science:

• Urology
• Medical Imaging
• Artificial Intelligence

Background:

• Cystoscopy is crucial for bladder cancer detection but has limitations.
• Artificial intelligence (AI) can enhance cystoscopy by identifying regions of interest (ROIs).
• Creating datasets for AI training in this field is challenging and time-consuming.

Purpose of the Study:

• To present CystoDS, a high-quality bladder imaging dataset for AI applications.
• To facilitate AI-driven detection of bladder cancer and cancer-mimicking benign lesions.
• To provide an AI-ready dataset for clinical decision support in cystoscopy.

Main Methods:

• Acquisition of 8,067 standard white light cystoscopy images from 160 patients.
• Labeling images with five classes and 22 subclasses, including segmentation data for 768 images.
• Detailed documentation of image acquisition, dataset structure, and technical validation.

Main Results:

• CystoDS is a comprehensive dataset ready for AI model training.
• The dataset supports the identification of both malignant and benign bladder lesions.
• Technical validation confirms the dataset's suitability for AI applications.

Conclusions:

• CystoDS addresses the need for high-quality data in AI-assisted cystoscopy.
• This dataset can advance AI development for improved bladder cancer diagnosis.
• The dataset supports AI in distinguishing cancerous from benign lesions during cystoscopy.