2018-06 - Undergraduate thesis project - Chongqing University
With the rapid development of medical technology, efficient and well-organized management of massive medical imaging data can effectively help healthcare workers eliminate the heavy workload of manual screening and diagnostic tasks. Therefore, establishing an efficient, reasonable, and accurate medical image retrieval system is the current need and expectation in the medical domain, which also reflects the demand for efficient medical data information retrieval and management. This research project, undertaken by China Medical University and Chongqing University, aims to study efficient medical image data retrieval technology based on deep learning and use image recognition to assist in discriminating secondary bleeding events in cerebral hemorrhage. The study focuses on the application of convolutional neural networks to cerebral hemorrhage CT image retrieval. The work in this paper includes the following:
- Exploring the pre-processing methods of cerebral hemorrhage CT images, including 1). image segmentation of the cerebral hemorrhage lesion area, and 2). image enhancement for the targets. Based on the grayscale characteristics of cerebral hemorrhage CT images, the K-means clustering method is chosen to perform the initial image segmentation of the cerebral hemorrhage lesion area, and the optimal k-value is selected using the elbow method to obtain the preliminary segmentation image. To further highlight the image features of the lesion area, the segmented cerebral hemorrhage CT image is binarized to enhance the targets.
- Using convolutional neural networks to extract typical cerebral hemorrhage lesion area features from the pre-processed cerebral hemorrhage CT images.
- Designing and implementing a web-based cerebral hemorrhage image retrieval system based on Django framework.
Model Design:
System Design:
Key words： convolutional neural network, Intracerebral hemorrhage CT image, image retrieval, K-means clustering method, Image binarization
Slides for presentation: https://drive.google.com/file/d/1wRwm8fQhdPXRxOZTHhLGAAtGqhw0tRCF/view?usp=sharing