Neuroevolutionary intelligent system to aid diagnosis of motor impairments in children
Lanzilotta, M. Ma, R. Accardi, M. RaviChandran, N. Zaernia, Amir Parisi, L.
Lanzilotta, M.
Ma, R.
Accardi, M.
RaviChandran, N.
Zaernia, Amir
Parisi, L.
Publication Date
2022-07
End of Embargo
Supervisor
Rights
Peer-Reviewed
Yes
Open Access status
closedAccess
Accepted for publication
2021-12
Institution
Department
Awarded
Embargo end date
Additional title
Abstract
An early detection of motor impairments in children is essential to improve self-care. Nevertheless, it may not be straightforward to conduct all required assessments physically in a clinic. Previous efforts in providing such a classification leveraged Machine Learning (ML)-based techniques on centralised data collected from activities of daily living (ADL); however, they used black-box optimisation methods and classification approaches that are not generalisable, hindering their application and impact in a clinical setting from an applied intelligence perspective. Thus, this study developed, applied, and validated a novel two-step neuroevolutionary optimisation approach to achieve earlier convergence in a user-centred and explainable manner, by applying early stopping based on three cross-entropy loss functions (for training, validation, and test) and optimise hyperparameters concurrently to ensure generalisation of the predictive performance by design. This method was applied to a neural network-based decision support system for aiding detection of motor impairments using three publicly available datasets on 190 children with motor impairments, including 483 socio-demographic and clinical self-care-related features in total. Such biomarkers were based on “The International Classification of Functioning, Disability, and Health for Children and Youth” (ICF-CY) and retrieved from the University California Irvine ML database, which were leveraged for the initial model training, validation, and testing. Moreover, to increase users’ confidence and drive adoption of the proposed applied intelligence-based tool, two additional publicly available datasets were used for validating the proposed two-step neuroevolutionary approach; such data include 34 and 39 socio-demographic and self-care-related clinical characteristics features as indicators of the functional independence of 50 further children with motor disability from a hospital in the city of Salvador, State of Bahia in Brazil. A novel, explainable and generalisable two-step neuroevolutionary approach guided early stopping based on three cross-entropy loss functions and optimised hyperparameters concurrently to improve generalisation in a neural network-based classifier. The ability of the proposed hybrid decision support system to detect motor impairments in the paediatric subjects considered was evaluated as compared to the best-performing ML-based classification systems on the same benchmark datasets to date. Whilst leveraging a transparent optimisation approach, results demonstrate its superior generalisation, quantified via the classification accuracy and the area under the receiver operating characteristic curve, being almost 6% higher than the most accurate and reliable ML-based system on the same benchmark dataset from published studies, thus supporting its potential for aiding classification of self-care problems in children with physical and motor disability in a clinical setting.
Version
No full-text in the repository
Citation
Lanzillotta M, Ma R, Accardi M et al (2022) Neuroevolutionary intelligent system to aid diagnosis of motor impairments in children. Applied Intelligence. 52: 10757-10767.
Link to publisher’s version
Link to published version
Link to Version of Record
Type
Article