Hussain, Arshian, Alibakhshi, Alireza, Abdi, Abas, Ufumaka, Isreal, Serej, Nasim Dadashi, Khalili, Hosseinali and Zolgharni, Massoud (2026) Uncertainty-weighted multi-task learning for early prediction of functional outcome and length of stay in traumatic brain injury. In: Artificial Intelligence in Healthcare. Springer, pp. 217-230.

PDF Uncertainty-Weighted Multi-task Learning for Early Prediction of Functional Outcome and Length of Stay in Traumatic Brain Injury.pdf - Published Version Restricted to Repository staff only Download (2MB)

Abstract

Accurate prediction of functional outcome and care trajectory after traumatic brain injury (TBI) is clinically important but challenging due to class imbalance and heterogeneous intermediate outcome states. This study evaluates how task-specific loss functions influence an uncertainty-weighted multi-task learning framework for simultaneous prediction of 6-month F-GOSE, ICU LOS, and hospital LOS from structured clinical registry data. Sparse categorical cross-entropy, weighted cross-entropy, focal loss, and alpha-weighted focal loss were compared under the same architecture and uncertainty-weighted optimisation scheme. Across repeated runs, loss choice affected task-wise and class-wise performance, with sparse categorical cross-entropy achieving the strongest F-GOSE accuracy and micro-AUC, while alpha-weighted focal loss achieved the highest F-GOSE macro F1-score. LOS performance was more similar across losses, although imbalance-aware losses provided marginal improvements for selected ICU and hospital LOS metrics. Intermediate outcome classes remained the most difficult to predict, highlighting persistent heterogeneity in disability-dependent functional outcome and moderate LOS groups.

Item Type:	Book Chapter or Section
Identifier:	10.1007/978-3-032-35387-0_16
Additional Information:	Book chapter of a conference paper presented in the International Conference on AI in Healthcare (AIiH), 28-28 August, London, United Kingdom.
Subjects:	Computing > Intelligent systems Medicine and health
Date Deposited:	24 Sep 2026
Dates:	Date Publication status 22 May 2026 Accepted 12 August 2026 Published Online
School, department or research centre:	School of Computing and Engineering THRIVE (The Centre for Translational Healthcare Research, Innovation, Vision, and Excellence)
URI:	https://repository.uwl.ac.uk/id/eprint/15320

Actions (admin access)