Summary: UCLA researchers in the Department of Electrical and Computer Engineering have developed a novel pre-processing computer algorithm for automated medical imaging.
Background: The integration of machine learning and artificial intelligence into medical imaging applications offers significant potential to automate the diagnosis of various disease and conditions. Despite these advancements, a significant challenge persists – managing the variability in imaging data, which can greatly affect the performance of diagnostic algorithms. This variability can arise from differences in sample preparation, discrepancies between imaging machines, human error, or variations in images captured by the same machine. Traditional preprocessing methods often struggle to correct for differences in lighting, contrast, and noise, which can obscure vital diagnostic information. In the realm of medical applications, where precision and reliability are crucial, these challenges are particularly pronounced. Misinterpretations caused by poor data quality can have severe consequences for patient care.
Current image enhancement techniques are not advanced enough to adapt to the unique characteristics of medical images, underscoring the need for more sophisticated methods. Physics-inspired computer vision (PhyCV) emerges as a promising solution to this problem. By reducing data variability and emphasizing critical diagnostic details, PhyCV can significantly enhance the consistency, accuracy, and reliability of machine learning models in medical image analysis. This approach not only holds the potential to improve patient outcomes, but also to improve the efficiency and cost effectiveness of general healthcare services.
Innovation: UCLA researchers have developed a novel preprocessing approach of medical data to improve accuracy of automated imaging tasks. This preprocessing algorithm transforms real-valued medical images into complex-valued signals, which may address variability issues that arise from differences in imaging practices, environments, machines, and storage. This method involves Physics-inspired Computer Vision (PhyCV) algorithms, which refine data before being fed into a machine learning model. This preprocessing step reduces non-semantic data variability, enhancing the quality and reliability of the medical data used during both training and inference stages of machine learning. This novel technique particularly benefits medical imaging applications such as pathology images, MRI, and X-rays, aiming to improve diagnostic accuracy and efficiency.
Potential Applications: • Enhanced imaging analysis for pathology images, MRIs, and X-rays • Automated disease detection including, but not limited, to cancer detection, neurological disorders, heart disease diagnostics, and vascular imaging • Precision medicine • Treatment response monitoring • Image-based biomarkers • Digital pathology • Telemedicine and remote analysis
Advantages: • Improved accuracy of automated medical image preprocessing • Highlighting the critical diagnostic features of medical imaging • Ability to handle variations in data from different imaging techniques, machines, or preparation • Adaptable to specific imaging requirements and types of medical data • Enhancement of machine learning efficacy, as well as ensuring a consistent and high-quality input data stream
Development-To-Date: Successful demonstrations and actual reduction to practice were performed on 3/26/21, 5/29/23, and 12/15/23.
Related Papers and Patents: 1. B. Jalali & M.H.Asghari,USPatent10275891B2(UC_2014_840_2_LA),“Phase Transform For Object And Shape Detection In Digital Images”. 2. Asghari, M. H., & Jalali, B. (2015). Edge detection in digital images using dispersive phase stretch transform. Journal of Biomedical Imaging, 2015, 6-6. 3. Jalali, B., & MacPhee, C. (2022). VEViD: Vision Enhancement via Virtual diffraction and coherent Detection. eLight, 2(1), 1-16. 4. B. Jalali and C. MacPhee, US63/380,92 (UC 2023-085) VEViD: Vision Enhancement via Virtual Diffraction and Coherent Detection,
Reference: UCLA Case No. 2024-177
Lead Inventor: Bahram Jalali