According to a recent report, artificial neural networks can improve the signal-to-background ratio in near-infrared imaging (NIR), sharpening blurred images into high resolution clinical pictures. The NIR-IIb probes that produce the deepest tissue penetration and sharpest images often have toxic elements, making them unfeasible for use in medical imaging for humans. Zhuoran Ma and colleagues investigated a way to improve the image contrast and clarity of FDA-approved biocompatible dyes, which typically detect fluorescence in the 700-1,000 nm range of the near infrared spectrum. Using approximately 2,800 in vivo images taken in mice, the authors trained, validated, and tested artificial neural networks with the intent to transform images produced in the shorter wavelengths into ones that resemble images taken in the NIR-IIb window of 1,500-1,700 nm. In a mouse injected with an FDA-approved dye, the neural network increased the signal-to-background ratio of lymph node images taken to greater than 100. The authors also compared the deep learning-enhanced imaging to actual NIR-IIb imaging of a mouse tumor. The enhanced image showed a 26.2 tumor-to-normal tissue signal ratio vs the actual image's 30.8. According to the authors, deep learning assisted imaging could improve diagnostics and image-guided surgery in the clinic.
###
Article #20-21446: "Deep learning for in vivo near-infrared imaging," by Zhuoran Ma, Feifei Wang, Weizhi Wang, Yeteng Zhong, and Hongjie Dai.
MEDIA CONTACT: Hongjie Dai, Stanford University, CA; tel: 1-650-723-4518; e-mail: <hdai1@stanford.edu>
Journal
Proceedings of the National Academy of Sciences