KidneyStoneTR: A Patient-Level Split Abdominal CT Dataset for Kidney Stone Detection

Dataset OverviewKidneyStoneTR is an abdominal computed tomography (CT) dataset consisting of axial CT slices collected from patients diagnosed with kidney stones and healthy control subjects. The dataset was developed as part of the Scientific and Technological Research Council of Turkey (TÜBİTAK) project 123E442, which focuses on deep learning–based kidney stone detection from abdominal CT images.The dataset was curated to enable reproducible evaluation of deep learning models by providing patient-level separation between training and testing data. It includes both a five-fold cross-validation setup and an independent hold-out test set reserved for final evaluation.All splits were generated at the patient level to avoid data leakage between training and testing subsets. The dataset supports binary classification of stone-positive and stone-negative CT slices and can be used for the development and benchmarking of computer-aided detection systems for kidney stone analysis in abdominal CT imaging.Dataset StructureThe dataset is organized into cross-validation splits and an independent hold-out test set.Dataset├── dataset_integrated_task│├── cross_validation│ ├── Training│ │ ├── Fold_1│ │ ├── Fold_2│ │ ├── Fold_3│ │ ├── Fold_4│ │ └── Fold_5│ ││ └── Test│ ├── Fold_1│ ├── Fold_2│ ├── Fold_3│ ├── Fold_4│ └── Fold_5│└── holdout├── Positive└── NegativeThe cross_validation directory contains a five-fold cross-validation split prepared at the patient level.The holdout directory contains an independent dataset reserved exclusively for final model evaluation.Dataset StatisticsCross-Validation SplitsFold-1TrainingStone+: 1443 images / 157 patientsStone−: 1424 images / 32 healthy subjectsTestStone+: 283 images / 39 patientsStone−: 302 images / 7 healthy subjectsFold-2TrainingStone+: 1415 images / 156 patientsStone−: 1352 images / 31 healthy subjectsTestStone+: 311 images / 40 patientsStone−: 374 images / 8 healthy subjectsFold-3TrainingStone+: 1341 images / 157 patientsStone−: 1359 images / 31 healthy subjectsTestStone+: 385 images / 39 patientsStone−: 367 images / 8 healthy subjectsFold-4TrainingStone+: 1364 images / 157 patientsStone−: 1386 images / 31 healthy subjectsTestStone+: 362 images / 39 patientsStone−: 340 images / 8 healthy subjectsFold-5TrainingStone+: 1341 images / 157 patientsStone−: 1383 images / 31 healthy subjectsTestStone+: 385 images / 39 patientsStone−: 343 images / 8 healthy subjectsIndependent Hold-Out Test SetStone+: 440 images / 45 patientsStone−: 292 images / 7 healthy subjectsSupported Research TaskThe dataset supports research on automated kidney stone detection from abdominal CT slices.Possible research applications include:• binary classification of stone-positive and stone-negative CT images• development of deep learning–based kidney stone detection systems• benchmarking computer-aided detection methods in abdominal CT imagingUsage NotesThe dataset is intended for academic research in medical image analysis and computer-aided diagnosis.The images are organized with patient-level separation between training and testing subsets to ensure fair model evaluation.The dataset should not be used for clinical decision-making without appropriate regulatory approval and validation.Citation and Usage PolicyIf you use this dataset in your research, please cite both the dataset and the following related publications:Öksüz, C., Narter, A., Ece, B., Koyun, M., Taşkent, İ., & Güllü, M. K. (2026). I2IReg–ClfNet: a cascaded multi-task deep learning framework for ROI-aware kidney stone detection in abdominal CT images. Biomedical Signal Processing and Control, 119, 109857. DOI: https://doi.org/10.1016/j.bspc.2026.109857Öksüz, C., Narter, A., Ece, B., Koyun, M., Taşkent, İ., & Güllü, M. K. (2025). A Hybrid CNN Framework for Kidney Stone Detection Using Transfer Learning and Feature Fusion. 33rd European Signal Processing Conference (EUSIPCO), pp. 1537–1541. DOI: 10.23919/EUSIPCO63237.2025.11226820Öksüz, C. (2026). Prospective pilot evaluation of a deep learning model for kidney stone detection on CT using a web-based workflow platform. International Urology and Nephrology. DOI: https://doi.org/10.1007/s11255-026-05057-9Öksüz, C. Slice-Level Classification of Kidney Organ Presence Using CNN–ViT Features: Toward Clinical Pre-Screening. Politeknik Dergisi (Advanced Online Publication).Dataset citation:Öksüz, C. (2026).KidneyStoneTR Dataset. DOI: 10.6084/m9.figshare.31156783