Dr. Oren Avram, a postdoctoral researcher at UCLA Computational Medicine, has led the development of SLIViT, an advanced deep-learning framework capable of analyzing 3D medical images like MRIs and CT scans with a precision comparable to medical specialists. The work, co-authored with Berkin Durmus, a UCLA PhD student, addresses the challenge of processing large volumetric medical datasets efficiently, making it applicable across various imaging modalities such as retinal scans and ultrasounds.
“SLIViT uses a unique architecture and also leverages knowledge from the more accessible 2D domain, enabling us to train it on moderately sized datasets, overcoming limitations typically seen in existing 3D models,” said Avram. This innovation enhances accuracy of state-of-the-art methods, requires significantly fewer training samples, and performs on par with experts, all while reducing diagnostic time drastically.
Their research advisor, Professor Eran Halperin, from the UCLA Samueli School of Engineering, praised the model’s ability to handle real-life conditions and smaller datasets, which potentially makes it a powerful tool for clinical applications. The team aims to expand SLIViT’s reach to predictive disease forecasting and address potential biases in AI models to ensure equitable healthcare outcomes.