LCA-Med: A lightweight cross-modal adaptive feature processing module for detecting imbalanced medical image
Xiang Li1, Long Lan2, Husam Lahza3
1College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China; Department of Intelligent Data Science, College of Computer Science and Technology, National University of Defense Technology, Changsha 410073, China; Department of Information Technology, Faculty of Computing and Information Technology, King Abdulaziz University, Jeddah 21589, Saudi Arabia; Chinese Medicine Guangdong Laboratory, Hengqin 519031, Guangdong, China.
This study introduces LCA-Med, a lightweight module for cross-modal medical image detection, improving accuracy despite data distribution discrepancies. The LCA-Med CNX method demonstrates superior performance on imbalanced datasets, advancing medical image analysis.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Data distribution discrepancy across medical image datasets hinders cross-domain adaptive detection accuracy.
- Existing methods struggle with heterogeneity in medical imaging modalities and data imbalance.
Purpose of the Study:
- To propose a novel lightweight cross-modal adaptive detection module (LCA-Med) for medical images.
- To develop an advanced cross-modal medical image adaptive detection method (LCA-Med CNX) and a new training paradigm.
- To enhance the accuracy and robustness of medical image detection, especially in imbalanced data scenarios.
Main Methods:
- Developed LCA-Med, a lightweight feature preprocessor extracting pathological information from varied image modalities guided by text prompts.
- Introduced LCA-Med CNX, a cross-modal medical image adaptive detection method.
- Proposed a novel training paradigm using generated dataset groups, an attention module, and a meta-heuristic algorithm.
Main Results:
- LCA-Med CNX achieved state-of-the-art performance on five out of six medical image datasets.
- The proposed method demonstrated superior performance compared to ten existing methods, particularly on imbalanced datasets.
- The lightweight LCA-Med module showed effective integration and feature extraction capabilities.
Conclusions:
- The proposed LCA-Med CNX method and training paradigm significantly improve cross-domain adaptive detection in medical imaging.
- The approach is effective in addressing data distribution discrepancies and data imbalance challenges.
- LCA-Med offers a versatile and efficient solution for feature extraction in multimodal medical image analysis.
