A two-stage deep learning framework for chest X-ray tampering detection and disease classification / Ali Hamdi

Public ISBD Titre : A two-stage deep learning framework for chest X-ray tampering detection and disease classification Type de document : document multimédia Auteurs : Ali Hamdi, Auteur ; Aymen Mouhouche, Auteur ; Younes Guellouma, Directeur de thèse Editeur : Laghouat : Université Amar Telidji - Département d'informatique Année de publication : 2026 Importance : 57 p. Accompagnement : 1 disque optique numérique (CD-ROM) Note générale : Option : Artificial intelligence and data science Langues : Anglais (eng) Mots-clés : Chest X-ray  Deep learning  Tampering detection  Disease classification  DenseNet-121  CBAM  NIH ChestX-ray14  Image authenticity Résumé : Making sure that medical images have not been altered is essential for any reliable AI diagnostic system. This thesis proposes a two-stage deep learning framework that first checks whether a chest X-ray has been tampered with, and then classifies it for 14 thoracic diseases. The system was built and tested on the NIH ChestX-ray14 dataset, which contains 112,120 radiographs from 30,805 patients. The work is structured around five experiments: (1) training a baseline disease classifier on clean images, (2) studying how tampering affects classifier performance, (3) training a dedicated tampering detector, (4) evaluating the full two-stage pipeline, and (5) analysing how the severity of tampering affects both components. Our classifier (DenseNet-121 + CBAM + Multi-Scale Feature Extractor) achieves a macro-AUC of 0.8303 on 15 classes. Tampering brings this down to 0.7928 (a 4.52% drop). The detector reaches AUC 0.9731 and accuracy 94.5%. The protected pipeline recovers 99.8% of clean baseline performance while rejecting 88.1% of tampered images. note de thèses : Mémoire de master en informatique A two-stage deep learning framework for chest X-ray tampering detection and disease classification [document multimédia] / Ali Hamdi, Auteur ; Aymen Mouhouche, Auteur ; Younes Guellouma, Directeur de thèse . - Laghouat : Université Amar Telidji - Département d'informatique, 2026 . - 57 p. + 1 disque optique numérique (CD-ROM). Option : Artificial intelligence and data science Langues : Anglais (eng) Mots-clés : Chest X-ray  Deep learning  Tampering detection  Disease classification  DenseNet-121  CBAM  NIH ChestX-ray14  Image authenticity Résumé : Making sure that medical images have not been altered is essential for any reliable AI diagnostic system. This thesis proposes a two-stage deep learning framework that first checks whether a chest X-ray has been tampered with, and then classifies it for 14 thoracic diseases. The system was built and tested on the NIH ChestX-ray14 dataset, which contains 112,120 radiographs from 30,805 patients. The work is structured around five experiments: (1) training a baseline disease classifier on clean images, (2) studying how tampering affects classifier performance, (3) training a dedicated tampering detector, (4) evaluating the full two-stage pipeline, and (5) analysing how the severity of tampering affects both components. Our classifier (DenseNet-121 + CBAM + Multi-Scale Feature Extractor) achieves a macro-AUC of 0.8303 on 15 classes. Tampering brings this down to 0.7928 (a 4.52% drop). The detector reaches AUC 0.9731 and accuracy 94.5%. The protected pipeline recovers 99.8% of clean baseline performance while rejecting 88.1% of tampered images. note de thèses : Mémoire de master en informatique

---

1 (1 - 1 / 1) Par page : 25 50 100 200