Fully Attention Convolutional Deep Neural Networks for Polyp Segmentation and Classification from Histological and Colonoscopic Images / Ines Mansour

Public ISBD Titre : Fully Attention Convolutional Deep Neural Networks for Polyp Segmentation and Classification from Histological and Colonoscopic Images Type de document : texte imprimé Auteurs : Ines Mansour, Auteur ; Nour Hamache, Auteur ; Abdelouaheb Moussaoui, Directeur de thèse Année de publication : 2023 Importance : 1 vol (177 f .) Format : 29cm Langues : Français (fre) Catégories : Thèses & Mémoires:Informatique Mots-clés : Colorectal cancer Colorectal polyps Index. décimale : 004 Informatique Résumé : Colorectal cancer (CRC) is a type of cancer that affects the colon or rectum, which are parts of the large intestine. It was the 3rd most commonly diagnosed cancer worldwide and the 2nd leading cause of cancer-related deaths after lung cancer in 2022. It can be prevented if glandular tissue (adenomatous polyps) is detected early. At the same time, Colonoscopy has been strongly recommended as a screening test for both early cancer and adenomatous polyps. However, its limitations include the high polyp miss rate for smaller (<10mm) or flat polyps, which are easily missed during visual inspection. Machine and Deep learning have emerged as promising tools in the field of cancer research and diagnosis and can be easily used in this context for the detection and segmentation of this type of cancer. In this work, we develop both Machine learning and Deep learning models that can segment and classify medical images to improve the diagnosis and treatment of colorectal cancer. We used Segmentation, and Classification techniques on both Colonoscopic and Histological images. For that, we propose two deep CNN convolutional neural network architectures from scratch,also we proposed CNN model with and attention mechanism such as CNN with CBAM, and we tuned 12 pre-trained models in histological images.in totle we proposed 18 models for histological images, and 6 others in colonoscopic images, that have been trained on large datasets, such as: → Machine Learning Classifiers such as ANN, KNN, SVM, RF, ABD, and XGB. → CNN Architectures without Attention mechanism : CNN Models from scratch. → CNN Architectures with Attention mechanism : CNN with CBAM. → Transfer Learning such as U-Net, DeepLabV3+, VGG, ResNet. → Transformers such as ViT (Vision Transformer): ViT B16, ViT B32. → Object Segmentation Methods such as SAM (Segment Anything Model). iii Our proposed models in Binary Classification on histological images obtained accuracy in a range [84.93%, 99,80%]. And our proposed models in Semantic Segmentation on colonoscopic images obtained an IoU score in the range [97.22%, 98.47%]. AI will revolutionize the medical field by enabling faster and more accurate diagnoses, personalized treatment plans, and improved patient outcomes. Côte titre : MAI/0712 Fully Attention Convolutional Deep Neural Networks for Polyp Segmentation and Classification from Histological and Colonoscopic Images [texte imprimé] / Ines Mansour, Auteur ; Nour Hamache, Auteur ; Abdelouaheb Moussaoui, Directeur de thèse . - 2023 . - 1 vol (177 f .) ; 29cm. Langues : Français (fre) Catégories : Thèses & Mémoires:Informatique Mots-clés : Colorectal cancer Colorectal polyps Index. décimale : 004 Informatique Résumé : Colorectal cancer (CRC) is a type of cancer that affects the colon or rectum, which are parts of the large intestine. It was the 3rd most commonly diagnosed cancer worldwide and the 2nd leading cause of cancer-related deaths after lung cancer in 2022. It can be prevented if glandular tissue (adenomatous polyps) is detected early. At the same time, Colonoscopy has been strongly recommended as a screening test for both early cancer and adenomatous polyps. However, its limitations include the high polyp miss rate for smaller (<10mm) or flat polyps, which are easily missed during visual inspection. Machine and Deep learning have emerged as promising tools in the field of cancer research and diagnosis and can be easily used in this context for the detection and segmentation of this type of cancer. In this work, we develop both Machine learning and Deep learning models that can segment and classify medical images to improve the diagnosis and treatment of colorectal cancer. We used Segmentation, and Classification techniques on both Colonoscopic and Histological images. For that, we propose two deep CNN convolutional neural network architectures from scratch,also we proposed CNN model with and attention mechanism such as CNN with CBAM, and we tuned 12 pre-trained models in histological images.in totle we proposed 18 models for histological images, and 6 others in colonoscopic images, that have been trained on large datasets, such as: → Machine Learning Classifiers such as ANN, KNN, SVM, RF, ABD, and XGB. → CNN Architectures without Attention mechanism : CNN Models from scratch. → CNN Architectures with Attention mechanism : CNN with CBAM. → Transfer Learning such as U-Net, DeepLabV3+, VGG, ResNet. → Transformers such as ViT (Vision Transformer): ViT B16, ViT B32. → Object Segmentation Methods such as SAM (Segment Anything Model). iii Our proposed models in Binary Classification on histological images obtained accuracy in a range [84.93%, 99,80%]. And our proposed models in Semantic Segmentation on colonoscopic images obtained an IoU score in the range [97.22%, 98.47%]. AI will revolutionize the medical field by enabling faster and more accurate diagnoses, personalized treatment plans, and improved patient outcomes. Côte titre : MAI/0712

Exemplaires (1)

Code-barres	Cote	Support	Localisation	Section	Disponibilité
MAI/0712	MAI/0712	Mémoire	Bibliothéque des sciences	Anglais	Disponible Disponible

---

1 (1 - 1 / 1)