MIT's deep learning technique could illuminate biological features in low-exposure images

An artificial intelligence (AI) technique developed by engineers at MIT in Cambridge, Massachusetts may be used to reveal transparent features in medical images taken with little to no light. The research was published online Dec. 12 in the journal Physical Review Letters.   

Led by George Barbastathis, PhD, professor of mechanical engineering at MIT, researchers trained a computer embedded with a deep neural network to recognize more than 10,000 grainy, dark patterns from objects photographed in very low lighting conditions.  

The computer was then shown a new grainy image not included in the training data set and learned how to reconstruct the transparent objects in the image that was obscured from the darkness. The method may be used to illuminate transparent features such as biological tissues and cells in images taken with very weak light. 

Additionally, the reconstructed images were more defined than a physics-informed reconstruction of the same patterns imaged in light that was more than 1,000-times brighter, according to the researchers.  

Overall, the findings point to using lower amounts of radiation and ultimately increasing patient safety. 

“What we’re doing here is, you can get the same image quality, but with a lower exposure to the patient,” Barbastathis said in a prepared statement. “And in biology, you can reduce the damage to biological specimens when you want to sample them.”