UC San Diego Develops AI That Reads Medical Images with Less Data to Improve Diagnosis Speed and Accuracy
Researchers at the University of California, San Diego, have created a new artificial intelligence tool that can learn to analyze medical images effectively using significantly fewer annotated samples than traditional methods. This innovation addresses a major challenge in medical imaging, where training deep learning systems usually requires large, expertly labeled datasets, which can be costly and time-consuming to produce. By reducing the amount of labeled data needed by up to twenty times, the AI system makes medical image segmentation more practical and accessible, especially in resource-limited clinical settings.
The AI Tool Mimics Radiologists’ Focus to Boost Diagnostic Efficiency
Led by electrical and computer engineering professor Pengtao Xie and doctoral student Li Zhang, the UC San Diego team developed the AI to perform precise segmentation of medical images, labeling each pixel to identify diseased versus healthy tissue. The system functions through a generative process that creates synthetic images tailored to improve segmentation learning. This continuous feedback loop tightly integrates data generation with model training, ensuring the AI develops accurate and relevant diagnostic capabilities. Tested on tasks including skin lesion detection, breast cancer ultrasound analysis, and placental vessel mapping, the AI consistently outperformed existing approaches, improving model accuracy by 10 to 20 percent while requiring far fewer real annotated images.
This breakthrough promises faster, more affordable diagnostic tools that replicate the focused attention of expert radiologists. It could dramatically enhance early detection and treatment planning for various medical conditions by making AI-powered imaging analysis viable in hospitals and clinics with limited data and resources. The research was published in Nature Communications and is supported by the National Science Foundation and National Institutes of Health.
UC San Diego’s AI innovation represents a significant step forward in medical imaging technology, aligning with ongoing efforts to integrate AI into healthcare to improve patient outcomes through smarter, efficient diagnostics.