Case Study - Presymptomatic Alzheimer’s Disease Detection
High-precision diagnostic model achieving >98% accuracy using optimized ensemble learning.
- Client
- Hotchkiss Brain Institute
- Year
- Service
- Diagnostic AI Optimization
Overview
The project aimed to detect Alzheimer's disease using hyperspectral imaging on blood samples. Various techniques were employed, including the MiniRocket algorithm for time series analysis, feature extraction to identify relevant data points, Gaussian smoothing to reduce noise, and MinMax scaling to normalize the data.
Several challenges were encountered, such as preventing data leakage to ensure the validity of the model and identifying outliers that could skew the results.
A range of deep learning models, including Gated Recurrent Units (GRU), Recurrent Neural Networks (RNN), Long Short-Term Memory networks (LSTM), and the HIVE-COTE 2.0, were tested. In addition, ensemble methods like XGBoost, LightGBM, and Random Forest were evaluated for their performance.
Among these methods, the Random Forest model showed particularly promising results, achieving an accuracy of over 98%.
What we did
- Gaussian Smoothing
- GRU
- RNN
- LSTM
- HIVE-COTE 2.0
- XGBoost
- LightGBM
- Random Forest
- Accuracy
- 98%
- Duration
- 3 Months