Retinal OCT Images Balanced for Training

OCT2026 is a reorganized and balanced version of the OCT2017 dataset containing retinal images captured via Optical Coherence Tomography. The dataset provides a 70/15/15 split for training, validation, and testing. Its origin and update history are not documented.

Use Cases

• Classify retinal OCT image types such as CNV, DME, DRUSEN, and NORMAL using convolutional neural networks.
• Train and validate image segmentation models on balanced splits of 70% training, 15% validation, and 15% test data.
• Benchmark model performance on a standardized split to compare against other medical image classification studies.
• Develop data augmentation strategies specific to retinal OCT image features to improve model generalization.

Strengths

• Provides a standardized 70/15/15 split for machine learning workflow.
• Balances classes from the original OCT2017 dataset to mitigate training bias.

Limitations

• The exact number of images, rows, and columns is unknown.
• The recency and temporal coverage of the source data are not specified.

Provenance

Source

Derived from the OCT2017 dataset.

Collection Method

Reorganized and balanced from the original collection.

Time Range

null

Freshness

null

Geography

null