End-to-End Deep Learning Vector Autoregressive Prognostic Models to Predict Disease Progression with Uneven Time Intervals



Bridge, Joshua, Harding, Simon ORCID: 0000-0003-4676-1158 and Zheng, Yalin ORCID: 0000-0002-7873-0922
(2021) End-to-End Deep Learning Vector Autoregressive Prognostic Models to Predict Disease Progression with Uneven Time Intervals. .

[img] Text
MIUA.docx - Author Accepted Manuscript

Download (2MB)

Abstract

We propose an end-to-end deep learning method combining implicit feature extraction and an autoregressive model to predict the future course of a disease or condition. By merging the feature extraction and autoregression into one deep learning model, we can simultaneously train both models together. Our novel approach begins by fine-tuning a pretrained convolutional neural network to extract features from previously obtained images of patients. A trainable autoregression mechanism then predicts the features of the future image and a fully connected layer gives a prognosis based on the predicted features. We utilize a novel time interval scaling, allowing the model to account for uneven time intervals and allowing us to choose the final time point that we wish to predict. Experiments on the Age-Related Eye Disease Study give a testing area under the receiver operating characteristic curve, sensitivity, and specificity of 0.966 (95% CI: 0.947, 0.984), 0.878 (0.810, 0.945), and 0.930 (0.914, 0.947), respectively. This shows that the model can predict progression with good performance.

Item Type: Conference or Workshop Item (Unspecified)
Uncontrolled Keywords: 4 Detection, screening and diagnosis, 4.1 Discovery and preclinical testing of markers and technologies
Divisions: Faculty of Health and Life Sciences
Faculty of Health and Life Sciences > Institute of Life Courses and Medical Sciences
Depositing User: Symplectic Admin
Date Deposited: 20 Oct 2021 10:26
Last Modified: 15 Mar 2024 08:39
DOI: 10.1007/978-3-030-80432-9_38
Related URLs:
URI: https://livrepository.liverpool.ac.uk/id/eprint/3140962