TransLPQPHOG: a new exemplar feature extraction model for intertrochanteric hip fracture detection
Rattachement africain : tr. Niveau de preuve : code pays fourni par la source.
Le résumé fourni par la source
Automated detection of intertrochanteric hip fractures from radiographs is constrained by the limited size of available datasets and the poor interpretability of many learning-based models. Here we develop TransLPQPHOG, a compact feature-engineering framework that integrates fixed local image patches, local phase quantization, pyramid histograms of oriented gradients, cumulative weighted iterative neighbourhood component analysis and multilayer perceptron classification. Each 224× 224-pixel radiograph was partitioned into a 14 × 14 grid of non-overlapping 16 × 16-pixel patches. This representation generated 83,104 features, of which 659 were retained for classification. We evaluated the framework in a retrospective single-centre cohort of 470 anteroposterior radiographs from 470 patients, comprising 226 intertrochanteric fracture cases and 244 controls. Image-level tenfold cross-validation was patient-disjoint because each patient contributed only one radiograph. Within every outer fold, min–max normalization, CWINCA ranking, feature-subset selection and MLP fitting were performed using the training partition only; the held-out fold was used only for testing. TransLPQPHOG achieved an accuracy of 90.85% (95% confidence interval, 87.90–93.14%), a sensitivity of 91.59%, a specificity of 90.16% and an area under the receiver operating characteristic curve of 0.9615. Under the same validation protocol, whole-image LPQ–PHOG achieved 72.77% accuracy, whereas patch-based configurations achieved accuracies ranging from 77.87% to 88.30%. The complete framework therefore provided the highest classification performance. Shapley analysis further showed that PHOG features accounted for 89.5% of the total absolute feature contribution, indicating that local gradient structure was the principal source of discriminative information. These findings demonstrate that interpretable patch-based feature engineering can support accurate internal classification of intertrochanteric hip fractures in a limited-data setting. However, the results do not establish clinical generalizability, and independent multicenter and prospective validation is required before clinical application.
Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.
Le contrôle bibliographique ouvert
DOI retrouvé dans Crossref DOI retrouvé ; titre concordant.
- Titre Crossref
- TransLPQPHOG: a new exemplar feature extraction model for intertrochanteric hip fracture detection
- Date Crossref
- 15/09/2026
- Éditeur
- Springer Science and Business Media LLC
- Type
- journal-article
Ce recoupement confirme des métadonnées liées au DOI. Il ne confirme ni la méthode ni les conclusions de l’étude, et il ne compte pas comme une seconde source scientifique indépendante.
Où se fait cette recherche
-
Elazığ Eğitim ve Araştırma Hastanesi pays non établi dans la noticeÉtablissement de santé
-
Ankara Yıldırım Beyazıt University pays non établi dans la noticeUniversité ou école supérieure
-
Erzurum Technical University pays non établi dans la noticeUniversité ou école supérieure
-
Fırat University pays non établi dans la noticeUniversité ou école supérieure
-
Fethi Sekin City Hospital Department of Orthopedics and Traumatology pays non établi dans la noticeÉtablissement de santé
-
Faculty of Medicine Department of Orthopedics and Traumatology pays non établi dans la noticeUniversité ou école supérieure
-
Ankara Yildirim Beyazit University Department of Electronics and Automation pays non établi dans la noticeUniversité ou école supérieure
-
College of Engineering Department of Computer Engineering pays non établi dans la noticeUniversité ou école supérieure
-
Technology Faculty Department of Digital Forensics Engineering pays non établi dans la noticeUniversité ou école supérieure
Elazığ Eğitim ve Araştırma Hastanesi, Ankara Yıldırım Beyazıt University et Erzurum Technical University, avec 6 autres affiliations.
Une affiliation ne permet pas de déduire la nationalité d’un auteur.