|
In this work, we show that deep learning and SLIM can be combined to quickly deliver superior results in tissue screening applications. This concept of combining QPI label-free data with AI with the purpose of extracting molecular specificity has been recently introduced by our laboratory as phase imaging with computational specificity (PICS) [Nat. Comm., in press]. Training on ten thousand SLIM images of piglet brain tissue with the 71-layer transfer learning model Xception, we created a two-parameter classification to differentiate the gestational size: either appropriate for gestational age (AGA) or small for gestational age (SGA), and diet: either an experimental regimen high in hydrolyzed fats or a control diet, with an accuracy of 80% and 81%, respectively, and a four-parameter classification (diet and size) with 62% accuracy. These results are significant, as it would otherwise be impossible for a trained histopathologist to distinguish such discrepancies.
|
