Scholarlink

<p>Identifying Alzheimer’s disease stages accurately is crucial for timely detection and immediate and effective treatment regimens. Although existing Alzheimer’s stage classification achieved better performance, it was not conducted to achieve an accuracy-complexity trade-off. This research work aims to create a highly accurate but lightweight Alzheimer’s disease stage classification system. With the Augmented Alzheimer MRI (Magnetic Resonance Imaging) Dataset V2, the proposed hybrid framework integrates deep features extracted from pre-trained Convolutional Neural Networks, such as InceptionV3, DenseNet121, and MobileNetV2, with the four-machine learning (ML) classifiers, including AdaBoost, Random Forest (RF), Support Vector Machine (SVM), and K-nearest Neighbors (KNN). Two hybrid setups are proposed, where the first is a dual-CNN with a single ML classifier, and the other is a single CNN with an ensemble of four ML classifiers. While the highest classification accuracy is achieved by DenseNet121 &amp; InceptionV3–KNN hybrid model, the DenseNet121 with an ensemble of four ML classifiers hybrid model yielded the best trade-off in classification performance and computational complexity. The hybrid model with DenseNet121 in Tensorflow Lite (TFLite) and ensembled four Machine Learning classifiers attained Accuracy, F1-Score, False Positive Rate, and False Negative Rate of 97.64%, 96.69%, 0.85%, and 2.68%, with the model size and Mean Inference time of 15.35 Megabytes (MB) and 235 milliseconds. The proposed hybrid model with Tensorflow Lite (TFLite) feature extractor is also tested on the Raspberry Pi Model B device to confirm its suitability on edge devices.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Building an Efficient Multi-stage Alzheimer’s Disease Classification System Using Computationally Effective Hybrid and Ensemble Models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月14日

查看全文
  • Muppavarapu Chethan,
  • P. Sabarinathan,
  • R. Arumuga Arun,
  • R. S. Rampriya,
  • Rajkumar Soundrapandiyan

摘要

Identifying Alzheimer’s disease stages accurately is crucial for timely detection and immediate and effective treatment regimens. Although existing Alzheimer’s stage classification achieved better performance, it was not conducted to achieve an accuracy-complexity trade-off. This research work aims to create a highly accurate but lightweight Alzheimer’s disease stage classification system. With the Augmented Alzheimer MRI (Magnetic Resonance Imaging) Dataset V2, the proposed hybrid framework integrates deep features extracted from pre-trained Convolutional Neural Networks, such as InceptionV3, DenseNet121, and MobileNetV2, with the four-machine learning (ML) classifiers, including AdaBoost, Random Forest (RF), Support Vector Machine (SVM), and K-nearest Neighbors (KNN). Two hybrid setups are proposed, where the first is a dual-CNN with a single ML classifier, and the other is a single CNN with an ensemble of four ML classifiers. While the highest classification accuracy is achieved by DenseNet121 & InceptionV3–KNN hybrid model, the DenseNet121 with an ensemble of four ML classifiers hybrid model yielded the best trade-off in classification performance and computational complexity. The hybrid model with DenseNet121 in Tensorflow Lite (TFLite) and ensembled four Machine Learning classifiers attained Accuracy, F1-Score, False Positive Rate, and False Negative Rate of 97.64%, 96.69%, 0.85%, and 2.68%, with the model size and Mean Inference time of 15.35 Megabytes (MB) and 235 milliseconds. The proposed hybrid model with Tensorflow Lite (TFLite) feature extractor is also tested on the Raspberry Pi Model B device to confirm its suitability on edge devices.