ACE Dataset

ACE (Automated Cyst Evaluation) is a curated collection of 2,266 Optical Coherence Tomography (OCT) B-scans obtained from 40 patients diagnosed with Diabetic Macular Edema (DME). It is specifically designed to support the development and validation of deep learning models for two primary tasks: segmentation of intraretinal cysts and clinical grading of disease severity. Key Features of the Dataset: Dual Annotation: Each B-scan is paired with a high-resolution binary mask for precise cyst delineation and a clinical grade assigned by expert ophthalmologists. Clinical Standards: Grading follows the ESASO (European School for Advanced Studies in Ophthalmology) criteria, categorizing scans into four severity levels: Class 0 (Absent), Class 1 (Mild), Class 2 (Moderate), and Class 3 (Severe). Patient-Level Structure: To prevent overfitting to specific anatomical features and ensure model generalizability, the data is organized with an explicit patient-level subdivision. Imaging Specifications: Scans were acquired using a DRI Triton Swept-Source OCT device at a resolution of 1024x992 pixels, then resized to a uniform 512x512 pixels for computational consistency. ACE serves as a non-trivial benchmark for the scientific community, characterized by significant morphological heterogeneity and clinically representative class imbalance.