| Titre : | Semi Supervised Learning For Medical Image |
| Auteurs : | Lahcene Mamen, Auteur ; Asma Ammari, Directeur de thèse |
| Editeur : | Biskra [Algérie] : Faculté des Sciences Exactes et des Sciences de la Nature et de la Vie, Université Mohamed Khider, 2024 |
| Format : | 1 vol. (94 p.) / ill., couv. ill. en coul / 30 cm |
| Langues: | Anglais |
| Mots-clés: | SSL, Medical Imaging, DL, Pneumonia Classification, Diffusion Denoising Probabilistic Models, ConvNeXt, AI in Healthcare, Respiratory Diseases. |
| Résumé : |
Since the inception of medical imaging technologies, significant advancements have continually reshaped the landscape of healthcare diagnostics and treatment planning. In recent years, the integration of Artificial Intelligence (AI), particularly Deep Learning (DL), has propelled medical imaging to new heights. Despite these advancements, the accurate classification of medical images remains challenging, particularly in the presence of limited labeled data and diverse pathologies. This thesis addresses these challenges by proposing a Semi-Supervised Learning (SSL) approach to enhance the classification accuracy of medical images, with a focus on respiratory diseases such as pneumonia. In this thesis, a comprehensive approach is implemented utilizing Denoising Diffusion Probabilistic Models (DDPM) and ConvNeXt architectures to overcome the limitations of existing methods. The DDPM is employed for generating high-fidelity synthetic medical images, which are then used to augment the training dataset. Concurrently, the ConvNeXt architecture is utilized for robust pseudo-labeling and classification, leveraging both labeled and unlabeled data to improve model performance. The proposed method is tested on a comprehensive dataset of chest X-radiation (X-ray) images, demonstrating significant improvements in the classification accuracy of pneumonia, even in scenarios with limited labeled data. The approach not only addresses the class imbalance issues but also enhances the model’s ability to generalize across different patient demographics and imaging conditions. Furthermore, the practical applications of this method are explored through case studies involving the classification of pneumonia in various patient demographics and imaging conditions. The results indicate that the proposed approach achieves high precision in identifying and classifying pneumonia, outperforming traditional methods and other state-of-the-art models. The effectiveness of the proposed SSL framework highlights its potential for broader applications in medical imaging, including the diagnosis of other respiratory conditions and the integration into real-time clinical workflows. This study sets the stage for future research aimed at tackling more complex medical imaging challenges, thereby enhancing patient care and diagnostic accuracy. |
| Sommaire : |
Acknowledgements I Abstract II R´esum´e III Arabic abstract IV List of Figures IX List of Tables XI List of Acronyms XII General Introduction 1 1 Medical Background 4 1.1 Introduction . . . . . . . . . . . . . . . . 4 1.2 Medical Imaging . . . . . . . . . . . 4 1.2.1 Importance of Medical Imaging . . . . . . . . . 4 1.2.2 Advances in Medical Imaging Technology . . . . . . 1.3 Anatomy and Physiology of the Respiratory System . . . . 7 1.3.1 Structure and Function of the Lungs . . . . . . . . 7 1.3.2 The Role of Alveoli in Gas Exchange . . .. . 8 1.3.3 Respiratory Mechanics . . . . . . . . . . . . . . . . . 8 1.4 Respiratory Diseases and Their Impact . . . . . . . . . . .. . . 9 1.4.1 Overview of Common Respiratory Diseases . . . . . . . 9V 1.4.2 Impact of Pneumonia on Alveolar Function . . . . . . . . 10 1.4.3 Other Respiratory Infections and Conditions . . . . . . .. 11 1.5 Diagnostic Techniques in Respiratory Medicine . . . . . . . . . . 13 1.5.1 Diagnosis of Pneumonia . . . . . . . . . . . . . . 13 1.5.2 Presentation of X-ray Images . . . . . . . . . . . . 14 1.6 Advanced Imaging Techniques . . . . . . . . . . . . . . . 17 1.6.1 Computed Tomography (CT) Scans . . . . . . 17 1.6.2 Magnetic Resonance Imaging (MRI) . . . . . . . . . . . . 18 1.6.3 Positron Emission Tomography (PET) . . . . .. 19 1.7 Challenges in Medical Imaging-based Diagnosis . . . . . .. . 21 1.7.1 Limitations of Current Diagnostic Methods . . . . . . . . 21 1.7.2 The Growing Demand for Improved Diagnostic Tools . . . 22 1.8 The Role of AI in Medical Imaging . . . . . . . . . . . 24 1.8.1 Introduction to AI in Medicine . . . . . . . .. . 24 1.8.2 Addressing Current Challenges with AI . . . . . . . 24 1.9 Conclusion . . . . . . . . . . . 25 2 Theoretical and Technical Background 28 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 28 2.2 Overview of AI in Medical Imaging . . 28 2.2.1 Definition and Scope . . . . . . . . . . . 28 2.2.2 Impact on Radiology . . . . . . . . . . . . . . . . 29 2.2.3 Machine Learning in Medical Imaging . . . . . . 30 2.2.4 Deep Learning in Medical Imaging . . . . . 32 2.3 AI Learning Strategies . . . . . . . . . . . . . . . .. . . . . 34 2.3.1 Supervised Learning . . . . . . . . . . . . . . .4 2.3.2 Unsupervised Learning . . . . . . . . . . . . . . . .. 35 2.3.3 Reinforcement Learning . . . . . . . . . . . . . . . . 35 2.3.4 Hybrid Learning Problems . . . 35 2.3.5 Statistical Inference in Learning . . . . . . . . 36 2.3.6 Learning Techniques . . . . . . . . . . . . . . . .36VI 2.4 Semi-Supervised approaches for Medical Imagi . . 38 2.4.1 Overview and Importance . . . . . . . . . . . . . . .. . 38 2.4.2 Challenges related to Labeled Data Acquisition . . . . . 39 2.4.3 Benefits of SSL for Medical Image Classification . . . . .. . 39 2.5 Review of SSL-based methods . . . . . . . . . 40 2.5.1 Consistency-Based Methods . . . . . . . . .. . 40 2.5.2 Graph-Based Methods . . . . . . . . . . . . . . . . . . 40 2.5.3 Adversarial Methods . . . . . . . . . . . .. . . 41 2.5.4 Other Methods . . . . . . . . . . . 43 2.6 Case Study: Pneumonia Detection . . . . . . . . . . . . . . . . . . 43 2.6.1 Overview of Medical Imaging for Pneumonia Diagnosis . 43 2.6.2 Recent Work and Advancements . . . . . . . . . . . . . 44 2.6.3 Summary of Recent Studies . . . . . . . . . . . . . 46 2.7 Technical Background . . . . . . . . . . . . . . . . . . . 47 2.7.1 Diffusion Models for Image Generation . . . . . . . .. . 47 2.7.2 ConvNext for Pseudo-Labeling and Classification . . . . 50 2.7.3 Approaches for Image Synthesis and Classification . . . . . 53 2.8 Conclusion . . . . . . . . . . . . . . 54 3 Methodology and Implementation 56 3.1 Introduction . . . . . . . . . . . . . . . . . . . 56 3.2 Dataset Description . . . . . . . . . . . . . . . . 57 3.2.1 Data Splitting and Usage . . . . . . . . . . . . . 57 3.2.2 Dataset Configuration for DDPM Models . . . . . . . 57 3.2.3 Dataset Configuration for ConvNext Models . . . . . . .. 57 3.2.4 Models and Configurations . . . . . . . . . . . 58 3.3 Detailed System Architecture . . . . . . . . . . . . . . . 58 3.4 Detailed Design and Implementation . . . . . . . . . . 59 3.4.1 DDPM Generation . . . . . . . . . . . 60 3.4.2 ConvNext for Classification . . . . . . . . . . . .. . 64 3.4.3 Integration of Semi-Supervised Learning . . . . .. . . . . . . . . 64 3.5 Data Flow and Processing . . . . . . . . . . . . . . . . . . . 66 3.6 Design Challenges and Considerations .. . . . . 67 3.7 Implementation and Evaluation . . . . . . . . . . . . . 67 3.7.1 Hardware and Software Requirements . . . . .. . 67 3.7.2 Configuration Settings . . . . . . . . . . . . . . . . . . 68 3.7.3 Integration Testing . . . . . . . . . . . . . . . . . . . . 69 3.8 Results and Analysis . . . . . . . . . . . . . . . . . 69 3.8.1 Presentation of Results . . . . . . . . . . . . . . . . 69 3.8.2 Training and Validation Metrics . . . . . . 69 3.8.3 Empirical Result . . . . . . . . . . . . . . . . 71 3.9 Performance Comparison . . . . . . . . . . . . . . 75 3.11 Conclusion . . . . . . . . . . . . . . . . 76 Conclusion and Perspectives 77 Bibliography 79 |
| Type de document : | Mémoire master |
Disponibilité (1)
| Cote | Support | Localisation | Statut |
|---|---|---|---|
| MINF/889 | Mémoire master | bibliothèque sciences exactes | Consultable |
