This repository introduced a hardware-friendly 1-D deep learning neural network to analyze average fluorescence lifetime from mono- and bi-exponential decays. And a traing aware quantization was employed to make it more compact. The core modification is using adder-based convolution to make it hardware-friendly. Both synthetic and real data outperform most existing iterative, non-iterative, and 1-D CNN.
Figures below are GUI designed based on the neural network, showing phasor plot, intensity image, average lifetime images of synthetic and real data.
For more details about the implementation, please refer to the paper. If you find this work useful, please consider citing this article. A hardware version is being developed, stay tuned.
Citation: Zang Z, Xiao D, Quan W, Zinuo Li, Chen Y, & Li DDU, "Hardware Inspired Neural Network for Efficient Time-Resolved Biomedical Imaging," 44th IEEE EMBC 2022.