News

Article

UCLA Investigators Develop AI Model to Improve 3D Imaging for Disease Predictions

Author(s):

The AI model, SLIViT, improves accuracy and efficiency in interpreting 3D medical scans, outperforming traditional methods and reducing specialists' workload.

UCLA Investigators Develop AI Model to Improve 3D Imaging for Disease Predictions

Oren Avram, PhD

Credit: LinkedIn

A team of investigators from UCLA have developed a model with artificial intelligence (AI) that can accurately interpret 3D images and predict disease risk factors.1

Compared with 2D images, 3D images add depth but require more time, skill, and attention for experts to interpret. Trained clinical specialists take a few minutes to detect disease biomarkers, such as measuring the volume of an anatomical swelling, in a 3D retinal imagining scan that may consist of nearly 100 2D images.

“While there are many AI (artificial intelligence) methods for analyzing 2D biomedical imaging data, compiling and annotating large volumetric datasets that would be required for standard 3D models to exhaust AI’s full potential is infeasible with standard resources,” said lead investigator Oren Avram, PhD, a postdoctoral researcher at UCLA Computational Medicine, in a statement. “Several models exist, but their training efforts typically focus on a single imaging modality and a specific organ or disease.”2

The team developed the AI model, named SLviT for slice integration by vision transformer, that can convert a volumetric scan into 2D images to generate a single prediction.1 The mode includes a combination of 2 artificial intelligence components and a unique learning approach.

“SLIViT overcomes the training dataset size bottleneck by leveraging prior ‘medical knowledge’ from the more accessible 2D domain,” said Berkin Durmus, a UCLA PhD student and co-first author of the article, in the press release.2

Investigators evaluated the machine learning model on 8 tasks, including classification and regression across 6 datasets and 4 volumetric imaging modalities: computed tomography (CT), magnetic resonance imaging (MRI), optical coherence tomography (OCT), and ultrasound.1 Additionally, the developers studied the model with 3D retinal scans (OCT) to assess for disease risk biomarkers, ultrasound videos for heart function, 3D MRI scans for liver disease severity assessment, and 3D CT for chest nodule malignancy screening.

SLIViT outperformed domain-specific state-of-the-art models. The study showed it has clinical applicability potential and matches the accuracy of the manual work of clinical specialist but reduces their work by a factor of 5,000.

“And unlike other methods, SLIViT is flexible and robust enough to work with clinical datasets that are not always in perfect order,” Durmus said.2

The model also saves the time it takes to train specialists how to annotate a newly identified disease-related risk factor at scale in biomedical images. With a small dataset, a single trained clinician can annotate the new risk factor in a few days, quickening the annotation process.

Not only does SLIViT improve diagnostic efficiency and timeliness but it also provides a foundation mode to accelerate the development of future predictive modes.

Future research includes studying additional treatment modalities, investigating how SLIViT can be used for predictive disease forecasting to improve early diagnosis, and exploring ways to prevent systemic biases in AI modes to contribute to health disparities

“What thrilled me most was SLIViT’s remarkable performance under real-life conditions, particularly with low-number training datasets,” said SriniVas R. Sadda, MD, a professor of Ophthalmology at UCLA Health and the Artificial Intelligence & Imaging Research director at the Doheny Eye Institute. “SLIViT thrives with just hundreds – not thousands – of training samples for some tasks, giving it a substantial advantage over other standard 3D-based methods in almost every practical case related to 3D biomedical imaging annotation.”

References

  1. Avram, O., Durmus, B., Rakocz, N. et al. Accurate prediction of disease-risk factors from volumetric medical scans by a deep vision model pre-trained with 2D scans. Nat. Biomed. Eng (2024). https://doi.org/10.1038/s41551-024-01257-9
  2. New AI Model Efficiently Reaches Clinical-Expert-Level Accuracy in Complex Medical Scans. News Wise. September 28, 2024. https://www.newswise.com/articles/new-ai-model-efficiently-reaches-clinical-expert-level-accuracy-in-complex-medical-scans. Accessed October 1, 2024.


Related Videos
Thumbnail for schizophrenia special report around approval of Cobenfy.
Thumbnail for schizophrenia special report around approval of Cobenfy.
Thumbnail for schizophrenia special report around approval of Cobenfy.
Thumbnail for schizophrenia special report around approval of Cobenfy.
Thumbnail for schizophrenia special report around approval of Cobenfy.
Comparing Treatment Options for Psoriatic Arthritis with Philip Mease, MD
© 2024 MJH Life Sciences

All rights reserved.