Creative Commons source reading · Data quality and algorithm evaluation
FLAME: Federated Learning and Aggregated Multi-Model Ensemble for Multi-Class Alzheimer’s Disease Stage Classification from Structured Clinical Data
Karim Gasmi; Lassaad Ben Ammar; Moez Krichen; Ahod Alghuried
Originally published in Diagnostics (Basel) · 2026-06-29
From the article
Selected source text. JournalOS removes markup and shortens excerpts where indicated. The full article provides the complete context.
Abstract
Background/Objectives: The precise identification of Alzheimer’s disease (AD) stages through clinical data is crucial for early diagnosis and suitable therapy. This classification remains troublesome due to overlap in cognitive profiles across different phases of illness progression. This study presents a comprehensive and advanced diagnostic system, termed FLAME, featuring an enhanced federated learning architecture for privacy-preserving multi-institutional implementation. It provides a systematic review of machine learning (ML) and deep learning (DL) models for the classification of five stages of Alzheimer’s disease (AD). The models include cognitively normal (CN), subjective memory complaints (SMC), early mild cognitive impairment (EMCI), late mild cognitive impairment (LMCI), and Alzheimer’s disease (AD). Methods: Sixteen…
4. Results
This section presents the experimental results obtained on Alzheimer’s Disease Neuroimaging Initiative (ADNI) clinical dataset comprising N=1737 subjects and d=362 heterogeneous features, partitioned into a training set (ntrain=1390, Test_data =0) and a held-out test set (ntest=347, Test_data =1) using the predefined split indicator. Four complementary evaluation perspectives are reported in sequence: (Section 4.1) classical machine learning models; (Section 4.2) deep learning architectures; (Section 4.3) a Genetic Algorithm-optimised hybrid ensemble; and (Section 4.4) a federated learning framework integrating distributed deep learning with a centralised LightGBM component. All metrics are computed exclusively on the held-out test set.
4.2. Deep Learning Results
Nine deep learning architectures were trained on the same d=63 feature set and evaluated on the identical held-out test set (ntest=347) under the five-class formulation {CN, SMC, EMCI, LMCI, AD}. The architectures span four design paradigms: fully connected networks (DNN 3-layer, DNN 5-layer), residual tabular learning (ResNet-Tabular), convolutional feature extraction (CNN-1D), sequential modelling (LSTM, BiLSTM), hybrid convolutional-recurrent (CNN-LSTM), self-supervised reconstruction (AutoEncoder-Clf), and transformer-based attention (Tab-transformer). All models were trained under identical conditions: AdamW optimiser with weight decay λ=10−4, cosine annealing learning rate schedule (η0=10−3, Tmax=80 epochs), class-weighted cross-entropy loss to mitigate the class imbalance inherent in the five-class ADNI cohort, gradient…
Editorial reading guide
- What data created and tested the model, and was evaluation independent?
- What is the cost of a false positive, false negative or out-of-distribution decision?
