Collaborative Denoising Autoencoder for High Glycated Haemoglobin Prediction

Alhassan, Zakhriya; Budgen, David; Alessa, Ali; Alshammari, Riyad; Daghstani, Tahini; Al Moubayed, Noura

Abstract

A pioneering study is presented demonstrating that the presence of high glycated haemoglobin (HbA1c) levels in a patient’s blood can be reliably predicted from routinely collected clinical data. This paves the way for performing early detection of Type-2 Diabetes Mellitus (T2DM). This will save healthcare providers a major cost associated with the administration and assessment of clinical tests for HbA1c. A novel collaborative denoising autoencoder framework is used to address this challenge. The framework builds an independent denoising autoencoder model for the high and low HbA1c level, which extracts feature representations in the latent space. A baseline model using just three features: patient age together with triglycerides and glucose level achieves 76% F1-score with an SVM classifier. The collaborative denoising autoencoder uses 78 features and can predict HbA1c level with 81% F1-score.

Citation

Alhassan, Z., Budgen, D., Alessa, A., Alshammari, R., Daghstani, T., & Al Moubayed, N. (2019, September). Collaborative Denoising Autoencoder for High Glycated Haemoglobin Prediction. Presented at 28th International Conference on Artificial Neural Networks (ICANN2019), Munich, Germany

• Presentation Conference Type: Conference Paper (published)
• Conference Name: 28th International Conference on Artificial Neural Networks (ICANN2019)
• Start Date: Sep 17, 2019
• End Date: Sep 19, 2019
• Acceptance Date: Jun 25, 2019
• Online Publication Date: Sep 9, 2019
• Publication Date: Jan 1, 2019
• Deposit Date: Jul 2, 2019
• Publicly Available Date: Nov 13, 2019
• Print ISSN: 0302-9743
• Publisher: Springer Verlag
• Pages: 338-350
• Series Title: Lecture notes in computer science
• Series Number: 11731
• Book Title: Artificial neural networks and machine learning – ICANN 2019; 28th International Conference on Artificial Neural Networks, Munich, Germany, September 17–19, 2019 ; proceedings.
• ISBN: 9783030304928
• DOI: https://doi.org/10.1007/978-3-030-30493-5_34
• Public URL: https://durham-repository.worktribe.com/output/1142524