Research on Ensemble Model of Multi-Stage Convolution Modules for Pathological Image Diagnosis（多段畳み込みモジュールの統合モデルを用いた病理画像診断に関する研究）
Type: dissertation
Abstract: Automated diagnosis of breast cancer (BC) pathological images has become a prominent research topic in medical imaging-based artificial intelligence in recent years.Pathological images inherently exhibit complexity, heterogeneity, and ambiguity.Coupled with the subtle differences between cancer cells and normal tissues in early-stage breast cancer diagnosis, automated pathological image diagnosis using deep learningparticularly convolutional neural networks (CNNs)-faces numerous challenges.Although existing classical CNN models have demonstrated strong performance, they frequently encounter problems such as premature loss of feature information and rapid convergence into a low-accuracy plateau when applied to pathological images due to small image sizes and deep network structures.Particularly, numerous studies tend to avoid transfer learning scenarios that involve fixed pre-trained parameters from natural images due to significant domain discrepancies between natural and pathological images.Such discrepancies often lead to pre-trained features being unresponsive to specific pathological features, thus severely restricting the effectiveness, flexibility, and ease of deployment for classical CNN models as backbone architectures in cross-domain adaptation (learning).To address these issues, this thesis proposes an ensemble model based on multi-stage convolution modules specifically designed for pathological image diagnosis.Taking sistently served as models for my own academic goals.Without his attentive supervision and encouragement, I would not have successfully completed my doctoral journey.
Author: 葛 睿剛
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7207988421