International Research Journal of Education and Technology

Medical image analysis, particularly for cancer diagnosis, faces challenges such as integrating information from multiple imaging modalities and dealing with noisy labels in medical data. To address these issues, we propose a hierarchical multimodal fusion architecture combined with weakly supervised learning. This framework effectively integrates data from diverse modalities, leveraging attention mechanisms and a hierarchical structure for efficient feature extraction and integration. The weakly supervised component enhances resilience to noisy labels, improving classification accuracy despite imperfect annotations. Experimental results across cancer datasets demonstrate that this approach outperforms state-of-the- art methods, offering significant potential for clinical applications such as early cancer detection, diagnosis, and treatment planning. By delivering accurate and reliable predictions, this work aims to support informed healthcare decision-making and better patient outcomes.

Key Words: Medical Imaging, Multimodal Integration, Noisy Labels, Hierarchical Fusion, MRI, CT scans, X-rays, Machine Learning, Python, Advanced Learning Techniques, Diagnostic Accuracy, Clinical Decision-Making, Patient Outcomes.