Doctorat LMD SIC

Permanent URI for this collectionhttps://dspace.univ-tlemcen.dz/handle/112/25267

Browse

Now showing 1 - 5 of 5

Deep Learning-Based Automatic Diagnosis of Diseases
(University of Tlemcen, 2026-02-14) Mecifi, Youssera Zoukha
Skin cancer is a growing global health concern and remains a major global health challenge due to its aggressive nature and the complexity of early diagnosis. Early diagnosis is the secret to improved patient outcomes, especially when certain subtypes, such as nodular melanoma or atypical nevi, continue to pose classification difficulties even for experienced clinicians. The increasing worldwide incidence of skin cancer highlights the pressing need for accurate, accessible, and early diagnostic tools that can support clinical decision-making and improve patient outcomes. In recent years, artificial intelligence (and more specifically, deep learning) has emerged as a powerful ally in medical imaging. Within dermatology, deep convolutional neural networks (CNNs) have shown great promise in assisting with skin lesion analysis and improving diagnostic performance. The motivation behind this thesis stems from the goal of enhancing public health through the development of intelligent, efficient, and cost-effective classification tools that harness technological innovation. Contributions Synthesis Our work is aimed to improve the field of computer-aided dermatological diagnosis by exploring and optimizing deep learning-based classification systems for skin cancer, with specific focus on binary discrimination between malignant and benign lesions. Through a multifaceted approach grounded in real-world imaging challenges and modern neural network architectures, we addressed critical limitations in dataset quality, class imbalance, captured images configuration bias, and diagnostic variability. We took advantage of the SLICE-3D dataset, a recent dermoscopic image database revealed in the ISIC 2024 challenge. Among the core challenges inherent in this dataset; and indeed most clinical imaging databases; was the extreme class imbalance with malignant samples being an incredibly small minority. To offset this, we employed a DCGAN-based image synthesis pipeline that successfully augmented malignant cases to balance the dataset. The synthetic images, when validated by both visual quality and downstream classification performance, were found to be an effective method for improving data diversity without compromising quality. Yet another notable aspect of this thesis was the particular focus devoted to images’ lighting modality, which is usually given short shrift but has quite a high degree of influence in dermoscopic imaging especially after researching and finding only few AI studies in this domain, they were all purely medical researchers conducted with dermatologists and not AI models. So, with the assistance of metadata fields such as tbp_tile_type, we investigated this relatively unexplored area and separated and tested independently three lighting conditions: non-polarized white light (3D:white), Cross-polarized light (3D:XP), and an intermediate set Y.Z. MECIFI 73 General conclusion combination. This allowed us to explore systematically the effect of light on automated diagnostic model performance; a study that, as far as we know, had not been previously investigated with such extent of detailed comparison in Deep learning filed. Four pre-trained convolutional neural networks, DenseNet121, ResNet50, MobileNetV2, and EfficientNet-B0, were tested to perform binary classification on the preprocessed datasets. The experimental results consistently demonstrate that dermoscopic lighting conditions have a decisive influence on CNN-based skin cancer classification performance. Across all evaluated architectures, non-polarized (white-light) dermoscopy yielded the highest Recall, Specificity, and AUC, confirming its superiority for detecting malignant lesions while maintaining reliable benign exclusion. This finding is clinically significant, as high Recall is essential for screeningoriented diagnostic systems. Among the tested models, EfficientNet-B0 achieved the best overall balance between accuracy and discriminative performance, particularly under white-light illumination, while MobileNetV2 and ResNet50 exhibited exceptionally high Recall and Specificity, making them well suited for sensitivity-critical clinical scenarios. These results align with dermatological knowledge, as white-light dermoscopy enhances superficial epidermal and pigmentationrelated structures that are highly informative for malignancy discrimination. The key contributions of the thesis are: • Successful implementation of DCGAN to address class imbalance in our clinical dataset and crafting a transfer learning model that achieved a very high and impressive level of accuracy. • Benchmarking dermoscopic lighting conditions (XP vs. White Light) on DCNN performance. • Methodologically rigorous comparison of multiple architectures across balanced datasets, supported with good metrics. • Demonstrating that metadata-aware experimentation can produce more clinically valuable models. Overall, this work emphasizes the power of deep learning especially when combined with synthetic data generation and imaging metadata to improve early detection of skin cancer. Future Research Directions and Perspectives The findings and contributions of this thesis open several promising avenues for further research aimed at advancing deep learning methodologies for reliable and clinically applicable skin cancer detection: • Exploration of More Advanced Generative Models for Data Augmentation: While this thesis investigated DCGAN for synthetic image generation, future work Y.Z. MECIFI 74 General conclusion could explore other GANs like StyleGAN or diffusion-based generative models and hybrid architectures. These approaches may yield higher-fidelity lesion synthesis, more diverse feature representation, and improved generalization when integrated into training pipelines. • Multi-Modal and Cross-Domain Learning: Extending the analysis beyond dermoscopy images by incorporating additional modalities such as clinical photographs, histopathology, or metadata (e.g., patient age, lesion location) could significantly enhance diagnostic robustness. Research into multimodal fusion strategies and domain adaptation techniques would support broader generalizability across institutions and imaging devices. • Systematic Investigation of Lighting Modalities: This thesis highlighted the impact of lighting conditions (white light, cross-polarized, mixed) on model performance. Future work should expand this study by including standardized multi-center datasets, quantifying robustness under diverse acquisition protocols, and developing adaptive models capable of automatically normalizing or correcting lighting variations. • Explainability and Clinical Trustworthiness of Models: Future studies should place stronger emphasis on explainable AI (XAI) methods tailored to dermatology. Designing interpretable saliency maps, lesion-specific feature attribution, or prototype-based reasoning can help bridge the gap between model predictions and dermatologists’ diagnostic reasoning, thus enhancing trust in clinical settings. • Transferability and Federated Learning Across Institutions: To overcome data scarcity and privacy concerns, research into federated learning and privacy-preserving training approaches for skin lesion datasets is recommended. This would allow collaboration across hospitals without direct data sharing, while ensuring that models remain robust and clinically validated across diverse populations. • Robustness under Data Shifts and Longitudinal Generalization: Future investigations should examine how models behave under domain shifts such as population diversity, new lesion subtypes, or evolving imaging technologies. Research into continual learning and model adaptation will be crucial for ensuring that AI systems maintain high diagnostic accuracy in dynamic real-world clinical environments. Overall, this work emphasizes the power of deep learning, especially when combined with synthetic data generation and imaging metadata, to improve early detection of skin cancer. As Y.Z. MECIFI 75 List of Published Work encouraging as the results are, there is potential future work that can enhance diagnostic aid systems.
Towards An Expressive Model For Safe and Efficient Manipulation of Graph Databases
(University of Tlemcen, 2026-05-07) Boudaoud,Abdelkrim
During the past decade, we have witnessed a wide speed use of graph databases for modeling and analyzing highly interconnected data, the relational model quickly reaches its limit when it comes to query and analyze such data. Despite that, the relational model still dominates in the Web. The first motivation of this thesis is to bridge this gap by presenting a framework for mapping and efficiently querying relational data within a graph paradigm, with contributions spanning theoretical foundations, scalability solutions, and practical implementations. First, we establish a rigorous formal foundation by introducing a complete mapping process that maps any relational database, along with its schema, data, constraints, and operations, to an equivalent graph database. This process is the first to holistically guarantee critical mapping properties: information and semantic preservation to prevent loss of meaning; query and update preservation through algorithms that automatically convert SQL (relational query language) to Cypher (graph query language); and monotonicity for efficient incremental updates. Next, we synthesize these foundations into R2G-ETL, an end-to-end software tool that automates the complete mapping process from relational systems to graph systems. As the first tool to enforce all formal preservation properties, it provides a validated and practical pipeline for real-world adoption. The second contribution of this thesis deals with the computational complexity of graph pattern matching on large-scale data. As querying of graph data is based on the subgraph isomorphism, an NP-Complete problem, querying a large graph data may be time-consuming even with the more efficient systems like Neo4j. One of the possible solutions is to distribute the large graph data into separated machines, execute the query in each machine, collect, and merge the results. To our knowledge, no work has proposed a complete formalization of this problem. We then formalize a distributed querying framework for Cypher, the most widely used graph query language. This framework is centered on three algorithms, the first one handles user query fragmentation for distributed execution; the second one focuses on assembling the partial results returned from each fragment; while The core algorithm coordinates all these operations between the coordinator and the fragments, managing them in a parallel, multi-threaded manner. The distributed framework can integrate seamlessly within any graph database system such as Neo4j, overcoming the limitations of previous approaches.
Un système de recommandation social et sémantique sensible au contexte
(University of Tlemcen, 2021-09-18) El Yebdri Zeyneb ép. Boukli-Hacene
Le travail présenté dans ce manuscrit se situe dans le domaine des systèmes de recommandation sensible au contexte (CARS) qui vise à améliorer les systèmes de recommandation traditionnels (SR) en prenant en compte les informations du contexte lors de la prédiction. Cependant, Ces systèmes souffrent de certains défis, tels que le démarrage à froid et la rareté des données. De nouvelles méthodes sont proposées pour surmonter ces problèmes. Nous proposons dans notre travail, trois contributions. La première contribution consiste à pallier les limites de l’approche du découpage sensible au contexte (Context-aware splitting approach (CASA)) qui représente l’une des approches de pré-filtrage les plus efficaces du système de recommandation sensible au contexte. Nous proposons d’ajouter principalement des informations de confiance ainsi sémantique pour améliorer la qualité de prédiction. La deuxième contribution, consiste à proposer une approche hybride intitulée : Approche de post-filtrage sensible au contexte basée sur la confiance, qui utilise des déclarations de confiance comme une information riche, avec la méthode de compensation de contexte qui appartient à l’approche de post-filtrage contextuel. Les résultats des expérimentations révèlent que ces approches améliorent la pertinence des recommandations et surpasse les autres approches non contextuelles de la littérature en terme de précision. Une autre approche que nous proposons : Approche de modélisation contextuelle basée sur la sémantique et la confiance (ST-CAMA), qui combine les informations de confiance et de contexte en utilisant la pondération contextuelle. Aussi, afin de sélectionner uniquement les voisins de confiance et qui ont des intérêts communs à l’item cible pour l’utilisateur actif, nous proposons de construire des clusters se basant sur des similarités sémantiques enrichi sémantiquement via les données ouvertes liées (LOD).
Reconnaissance automatique des plasmocytes pour l’aide au diagnostic du Myélome
(University of Tlemcen, 2018-02-04) Baghli,Ismahan
La maladie du Myélome multiple appelée aussi maladie de Kahler, est par mis les maladies du cancer du sang. Ce type de cancer est caractérisé par la prolifération des globules blancs de type plasmocyte. Le diagnostic précoce de cette maladie, améliore le taux de survie du patient atteint. Le diagnostic manuel repose sur l’examination visuelle, par le pathologiste, des images mi croscopiques de la moelle osseuse pour chercher des signes de prolifération de cellules. Cette analyse nécessite des efforts considérables pour une détection fiable et précise, ce qui la rend à la fois lente et fastidieuse. Les systèmes automatiques basés sur l’identification et le comptage des globules blancs, peuvent donner des résultats plus précis que les méthodes ma nuelles. Ces systèmes sont basés principalement sur deux étapes essentielles : la segmentation des cellules, et la caractérisation et la classification des globules blancs. Dans le modèle automatique que nous proposons, les images micro scopiques de la moelle osseuse sont tout d’abord segmenetées par la trans formation watershed et ensuite les régions sont fusionnées par l’intégration de l’incertitude sur la couleur via la théorie de l’évidence. Ces globules blancs seg mentés sont caractérisés par les attributs de formes du noyau et de la cellule, et les attributs de textures couleurs du noyau et du cytoplasme, et au finale ces attributs sont introduits à trois classifieurs à apprentissage supervisé SVM (Support Vector Machine), Kppv (K plus proche voisins) et arbre de décision. Les résultats expérimentaux montrent que la reconnaissance du plasmocyte par le Kppv est prometteuse, atteingnant un taux de reconnaissance de 97% avec une spécificité de 100%, ce qui la rend insensible aux faux positifs.
Medical images indexation and annotation
(University of Tlemcen, 2020-11-17) Nedjar,Imane
Computer aided detection and diagnosis CADe/CADx systems, are an essential tools used by physicians to assist them in their daily clinical diagnosis. In cancers diseases, these systems have an important role to perform the early detection and diagnosis, this allows to provide early treatment before it will be too late. In this thesis, we present several methods to be uses in a computer aided diagnosis system in order to generate structured reports of liver lesions including cancer using Computed Tomography (CT) images. In addition, we propose different methods for computer aided detection of breast cancer, by treating breast density classification using mammography and breast lesion classification using histopathology images. At this context we present three distingue contributions, the first one is related to the annotation of liver CT images by using a medical ontology, in which we propose three methods. The second contribution is about breast density classification according to the standard Breast Imaging Reporting and Data System (BI-RADS). In addition to that, we propose an improved version of Synthetic Minority Over-Sampling Technique Algorithm (SMOTE) used to equilibrate the dataset. The last contribution is about breast lesions classification in the histopathology images. Precisely, we propose a method to distinct benignant and malignant lesions, as well to classify the normal cases, benign cases, in situ and invasive cancer cases.

Browse

Recent Submissions