On 1 May, Dr Shangran Qiu and colleagues published an article in Brain, detailing the development and validation of an AI-based framework for predicting the risk of developing Alzheimer’s disease (AD).
The last two decades have seen the rapid development of multimodal diagnostics for AD, involving biomarker tests, analyses of brain scans and genetics alongside more traditional cognitive assessments. Whilst offering the potential for highly precise diagnoses and risk estimation for AD, the accuracy of these multimodal tests currently depends on clinicians having the skills to interpret a diverse range of complex data sources. Artificial Intelligence (AI)-based approaches such as machine learning and neural networks could help clinicians rapidly perform in-depth analyses of multiple health data sources, enabling more uniform and accurate diagnosis of AD. However, there is a lack of external validation of AI-based algorithms, as well as a paucity of clear decision-making frameworks that clinicians can apply based on these algorithms.
To overcome these challenges, Dr Qiu and colleagues aimed to develop and validate a deep learning framework for creating high-resolution visualisations of AD risk by linking two types of neural networks: a fully convolutional network (FCN) and a multilayer perceptron (MLP). The framework was developed and validated using four case-control clinical datasets from participants with AD (cases) or cognitively unimpaired controls: the AD Neuroimaging Initiative (ADNI) dataset, the Australian Imaging, Biomarker and Lifestyle Flagship Study of Ageing (AIBL) dataset, the Framingham Heart Study (FHS) dataset, and the National Alzheimer’s Coordinating Centre (NACC) dataset.
The first step of the framework involved the random sampling of patches from MRI images in these datasets, leading to the creation of participant-specific disease probability maps in step 2. The third and final step of the framework resulted in an overall AD prediction, based on an integrated analysis of brain imaging data alongside factors such as age, gender and cognitive test scores. Validation tests of the framework showed that it displayed good predictive performance across the different clinical cohorts, despite between-cohort variations in MRI protocol, geographic location and recruitment criteria. Further studies are now required to test the accuracy of the AI framework in more diverse populations, with a range of comorbidities and potential neurodegenerative disease diagnoses.
The original article was published in Brain and can be found here: https://academic.oup.com/brain/advance-article/doi/10.1093/brain/awaa137/5827821#203273794
