Machine‐learning in motor neuron diseases: Prospects and pitfalls

Abstract

Although machine-learning (ML) approaches have been extensively utilized in neurodegenerative conditions, they can be challenging to implement in motor neuron diseases (MNDs) due to disease-specific characteristics. The potential of ML algorithms has been explored by academic amyotrophic lateral sclerosis (ALS) studies, but they have not been developed into viable clinical applications to date. ALS studies traditionally conduct "group-level" analyses to describe phenotype-or genotype-associated clinical traits, survival characteristics, progression rates, biomarker profiles, and imaging signatures [1-4]. These, although academically interesting, have limited utility

Hippocampal Metabolic Alterations in Amyotrophic Lateral Sclerosis: A Magnetic Resonance Spectroscopy Study

“…The limitations of structural imaging in ALS are seldom enunciated with sufficient candour, and these are particularly apparent when structural imaging is used in isolation without supporting fMRI and DTI data. Often, motor cortex changes are not detected by structural pipelines alone, and efforts to discriminate PLS and ALS patients based on T1-weighted and DTI data alone in machine-learning frameworks have been disappointing [ 97 , 98 ]. Accurate individual patient classification into diagnostic, phenotypic, or prognostic categories is an emerging field of ALS [ 98 , 99 , 100 ] and a multitude of promising initiatives have been reported using either clinical variables alone [ 101 , 102 , 103 ], imaging metrics [ 62 , 97 , 104 ], or both [ 105 ].…”

“…Often, motor cortex changes are not detected by structural pipelines alone, and efforts to discriminate PLS and ALS patients based on T1-weighted and DTI data alone in machine-learning frameworks have been disappointing [ 97 , 98 ]. Accurate individual patient classification into diagnostic, phenotypic, or prognostic categories is an emerging field of ALS [ 98 , 99 , 100 ] and a multitude of promising initiatives have been reported using either clinical variables alone [ 101 , 102 , 103 ], imaging metrics [ 62 , 97 , 104 ], or both [ 105 ]. Careful feature selection is indispensable for effective MRI-based machine-learning strategies, and most existing studies use solely structural and DTI data [ 106 , 107 ].…”

Thalamic pathology in frontotemporal dementia: Predilection for specific nuclei, phenotype‐specific signatures, clinical correlates, and practical relevance

“…MRI‐based classification models use discriminatory MRI features to categorize single‐subject MRI data into diagnostic groups. Feature selection in ALS–FTD spectrum disorders typically focuses on cortical gray matter thickness, volumes, and white matter metrics (Bede et al., 2022 ; Egger et al., 2020 ; Grollemund et al., 2019 ; Kim et al., 2019 ; McKenna, Murad, et al., 2022 ; Premi et al., 2016 ; Schuster et al., 2016 , 2017 ) rather than subcortical volumes; this is likely because subcortical volumes are considered as a whole instead of the inclusion of nucleus‐based metrics in the models. Thus, the addition of thalamic nuclei and thalamic radiation integrity metrics may improve the classification accuracy of such models (Bede et al., 2021 ).…”

Brainstem–cortex disconnection in amyotrophic lateral sclerosis: bulbar impairment, genotype associations, asymptomatic changes and biomarker opportunities

“…Machine-learning frameworks have been increasing applied to large ALS datasets [ 48 , 49 ] and feature importance analyses have invariably highlighted the role of cortical grey and white matter diffusivity measures [ 36 , 50 – 55 ]. To demonstrate the diagnostic utility of such models however, classification models need to be tested and validated on early-stage patients or patients soon after their diagnoses [ 56 ]. The accurate categorisation of late-stage or patients with considerable disability says relatively little about the practical utility of a particular model.…”