Medical Image Analysis Based on Deep Learning Methods

• Collected various medical image datasets, including cell and nuclei datasets, tumour and polyp datasets, and tissue and vessel datasets. Standardized the label format of the dataset and converted mask to NumPy format. Analysed datasets.
• Improved three prompts based on Segment Anything Model: point prompt, box prompt, text prompt.
• Improved SAM with automatically generated bounding box prompt. Implement YOLOv8 model in the object detection module to automatically generate bboxes for SAM’s box prompt input. The constructed YOLO_SAM model achieved good segmentation accuracy and improved the efficiency and applicability of medical segmentation tasks.
• Wrote the paper SPPNet: A Single-Point Prompt Network for Nuclei Image Segmentation, which has been accepted by MICCAI-MLMI 2023.
• Proposed and achieved three architectures for nuclei image segmentation: SPPNet, YOLO_SAM model, Mask GroundingDINO, and compared model performance and metrics on different medical image datasets. Summarized and organized all experimental results and completed the Final Year Report.