AI Tool for Early Dementia Prediction
Author Information
Author(s): Lee Liz Yuanxi, Vaghari Delshad, Burkhart Michael C., Tino Peter, Montagnese Marcella, Li Zhuoyu, Zühlsdorff Katharina, Giorgio Joseph de, Williams Guy, Chong Eddie, Chen Christopher, Underwood Benjamin R., Rittman Timothy, Kourtzi Zoe
Primary Institution: University of Cambridge
Hypothesis
Can a robust and interpretable predictive model improve early dementia diagnosis and prognosis using non-invasive data?
Conclusion
The predictive model accurately forecasts whether patients with mild cognitive impairment will progress to Alzheimer's disease, outperforming standard clinical markers.
Supporting Evidence
- The predictive model achieved an accuracy of 81.66% and an AUC of 0.84.
- It demonstrated sensitivity of 82.38% and specificity of 80.94%.
- The model was validated against longitudinal clinical outcomes.
- It reduced misdiagnosis compared to standard clinical markers.
- The model generalizes well across different memory clinics.
- It provides a continuous index of cognitive health trajectories.
- The PPM-derived prognostic index was significantly different across patient groups.
- The model's predictions were validated against independent real-world data.
Takeaway
Researchers created a smart tool that helps doctors figure out if someone with early memory problems will get worse, using simple tests instead of expensive scans.
Methodology
The study developed a predictive model using cognitive tests and MRI data from multiple centers to assess its ability to predict dementia progression.
Potential Biases
Potential biases may arise from class imbalance in the training data and the representativeness of the cohorts used.
Limitations
The model's generalizability may be limited by the diversity of the training and testing populations and the data collection methods used.
Participant Demographics
Participants included individuals from diverse clinical settings in the UK and Singapore, with varying demographics.
Statistical Information
P-Value
0.01
Confidence Interval
[1.36, 8.60]
Statistical Significance
p<0.01
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website