Toward a unified public x-ray dataset integrating multiple databases to advance complex fracture analysis [r-libre/4021]

Atitallah, Fatma; Thabet, Assem; Ayena, Johannes C et Mezghani, Neila (2026). Toward a unified public x-ray dataset integrating multiple databases to advance complex fracture analysis. Dans AIHealth 2026 : The Third International Conference on AI-Health (p. 75-81). Valencia, Spain. ISBN 978-1-68558-362-0

Fichier(s) associé(s) à ce document :

PDF - aihealth_2026_1_190_80159.pdf
Contenu du fichier : Version de l'éditeur

Télécharger

Catégorie de document :	Communications dans des actes de congrès/colloques
Évaluation par un comité de lecture :	Oui
Étape de publication :	Publié
Résumé :	Detecting and segmenting complex upper-limb frac- tures in X-ray imaging remains a persistent challenge due to subtle visual patterns, anatomical overlap, and heterogeneous acquisition conditions. While deep learning continues to ad- vance fracture analysis, its progress is limited by fragmented public datasets that differ in format, annotation standards, clinical focus, and image quality. This paper addresses this limitation by unifying three publicly accessible X-ray datasets (FracAtlas, Bone Fracture Detection, and GRAZPEDWRI-DX) into a single, standardized, high-quality resource dedicated to upper-limb fracture research. The proposed integration pipeline includes rigorous dataset selection, annotation harmonization, preprocessing, normalization, and multi-stage quality control to ensure consistency and clinical reliability. The resulting dataset provides extensive anatomical diversity, a wide spectrum of fracture types, and verified segmentation masks suitable for training and benchmarking deep learning models. Although class imbalance and uneven fracture representation persist, the unified dataset establishes a valuable foundation for developing robust, generalizable, and clinically meaningful Artificial Intelligence (AI) systems for fracture detection, localization, and segmenta- tion. Future extensions will focus on balancing fracture categories and evaluating state-of-the-art architectures on the proposed dataset.
Adresse de la version officielle :	https://www.thinkmind.org/library/AIHealth/AIHealt...
Déposant:	Ayena, Johannes
Responsable :	Johannes Ayena
Dépôt :	17 mars 2026 17:09
Dernière modification :	17 mars 2026 17:09

Actions (connexion requise)

RÉVISER