Table 1 Overview of current applications, challenges and prospects for machine learning and AI applications in five key areas of dementia research
From: Harnessing the potential of machine learning and artificial intelligence for dementia research
| Â | Current areas of machine learning and AI applications | Challenges and knowledge gaps | Prospects and future directions |
|---|---|---|---|
Genetics | Full genomic signal analysis [3, 4] Statistical fine mapping [6, 7] Identification of causal variants [6] | Effect of specific genetic variants [5] Relation of genetic variation to cellular changes [5] Mixed evidence for interaction of genetics with modifiable risk [21−24] | Utilisation of integrative data sets [124] Combining omics data to identify functional implications [125] Application of genetic risk to individuals [15] |
Experimental Medicine | Data-driven multimodal analysis [35, 36] Digital twin brain models link [33]structure, function and pathology | Translational gap from models to human disease biology [126] Lack of power in small, single modality studies [126] Poor reproducibility [127] | Efficient drug target discovery [128] Simulated ageing signatures [129] Digital brains for precision dementia research34 |
Drug discovery and Trials Optimisation | Intelligent drug target identification [16] Incorporation of multiple biomarker data [59] Natural language processing and text mining of electronic health records [68] | Heterogeneity of disease risk, severity and subtype [60, 61] Cost of longitudinal analysis [60] Restricted access to clinical trial data [64] | Enhanced identification of risk for trial recruitment [63] Utilising publicly available data and linked health records [16, 66] Multi institutional collaborative initiatives to share data [67] |
Neuroimaging | Automated feature extraction for diagnosis and prediction [85] Combining imaging modalities and biomarker data [86] Investigation of disease progression and biological mechanisms [61, 87] | Lack of clinical implementation [85] Poor interpretability is challenging for regulation [91] Sensitivity to bias in the training data [91] | Validation of existing models for clinical settings [90] Availability of large data sets and repositories [85] Strategic recruitment to improve real-world applicability [91] |
Prevention | Analysis of complex interactions in observational studies [113] Increased accuracy of polygenic risk and predictive models [15, 130] Validation of drug repurposing for dementia prevention [117, 118] | Inconsistent evidence for many potential risk factors [93, 95,96,97] Causal relationships poorly understood [98] Lack of statistical power [17] | Personalised dementia prevention interventions [122, 123] Deep learning for improved Mendelian randomisation [103, 105] Lifespan modelling to identify the optimal timing of a prevention intervention |