Temporal deep learning framework for retinopathy prediction in patients with type 1 diabetes

作者：

Highlights：

• We propose a deep learning framework for modeling irregular medical time series.

• We represent time as an input feature and as a learnable parameter of the neural network.

• We validate our framework on the HbA1C time series derived from a French database.

• We conduct a comparative study of Transformer-based and LSTM-based sequential models.

摘要

•We propose a deep learning framework for modeling irregular medical time series.•We represent time as an input feature and as a learnable parameter of the neural network.•We validate our framework on the HbA1C time series derived from a French database.•We conduct a comparative study of Transformer-based and LSTM-based sequential models.

论文关键词：Artificial intelligence,Deep learning,HbA1c,Time irregularity,Type 1 diabetes,Retinopathy

论文评审过程：Received 24 October 2021, Revised 17 September 2022, Accepted 21 September 2022, Available online 26 September 2022, Version of Record 14 October 2022.

论文官网地址：https://doi.org/10.1016/j.artmed.2022.102408