An Interpretable Machine Learning Model with Deep Learning-based Imaging Biomarkers for Diagnosis of Alzheimer’s Disease

Aug 3, 2026

Upload by Patrick

Machine learning enables automatic early diagnosis of Alzheimer’s Disease, yet many imaging-based models remain difficult to interpret because their decision mechanisms are unclear. The work introduces a hybrid framework that combines Explainable Boosting Machines, grounded in generalized additive modeling, with deep learning-based feature extraction for high-dimensional imaging biomarkers. The model yields feature-level importance for interpretation, and was validated on ADNI with strong accuracy and AUC on Alzheimer’s vs control, as well as external testing for Alzheimer’s vs SCD. Results show clear gains over image-volume EBM baselines and competitive performance against optimized CNNs.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages11

Reading Time28 min

AccessFree to download

Abstract

Introduction

Methods

Study population

Validation and performance evaluation

arXiv :2308 .07778v1 [ ee ss .IV] 15 Aug 2023 An Interpretable Machine Learning Model with Deep Learning-based Imaging Biomarkers for Diagnosis of Alzheimer’s Disease Wenjie Kang⋆1, Bo Li 1 , Janne M. Papma2 , Lize C. Jiskoot2 , Peter Paul De Deyn3 , Geert Jan Biessels4 , Jurgen A.H. R. Claassen5 , Huub A.M. Middelkoop6 ,7 , Wiesje M. van der Flier8 , Inez H.G.B. Ramakers9 , Stefan Klein 1 , Esther E. Bron 1 , for the Alzheimer’s Disease Neuroimaging Initiative, and on behalf of the Parelsnoer Neurodegenerative Diseases study group 1 Department of Radiology & Nuclear Medicine, Erasmus MC, Rotterdam, The Netherlands 2 Department of Neurology, Erasmus MC, Rotterdam, The Netherlands 3 Department of Neurology & Alzheimer Center, University Medical Center Groningen, Groningen, The Netherlands 4 Department of Neurology, UMC Utrecht Brain Center, University Medical Center Utrecht, Utrecht, The Netherlands 5 Radboud University Medical Center, Nijmegen, The Netherlands 6 Department of Neurology & Neuropsychology, Leiden University Medical Center, Leiden, The Netherlands 7 Institute of Psychology, Health, Medical and Neuropsychology Unit, Leiden University, The Netherlands 8 Amsterdam University Medical Center, location VUmc, Amsterdam, The Netherlands 9 Alzheimer Center Limburg, School for Mental Health and Neuroscience (MHeNS), Maastricht University Medical Center, Maastricht, The Netherlands Abstract. Machine learning methods have shown large potential for the automatic early diagnosis of Alzheimer’s Disease (AD) . However, some machine learning methods based on imaging data have poor interpretability because it is usually unclear how they make their decisions. Explainable Boosting Machines (EBMs) are interpretable machine learning models based on the statistical framework of generalized additive modeling, but have so far only been used for tabular data. Therefore, we propose a framework that combines the strength of EBM with highdimensional imaging data using deep learning-based feature extraction. The proposed framework is interpretable because it provides the importance of each feature. We validated the proposed framework on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset, achieving accuracy of 0.883 and area-under-the-curve (AUC) of 0.970 on AD and control classification. Furthermore, we validated the proposed framework on an external testing set, achieving accuracy of 0.778 and AUC of 0.887 on AD and subjective cognitive decline (SCD) classification. The proposed framework significantly outperformed an EBM model using vol- ⋆ [w.kang@erasmusmc.nl](w.kang@erasmusmc.nl) 2 W. Kang et al. ume biomarkers instead of deep learning-based features, as well as an end-to-end convolutional neural network (CNN) with optimized architecture. Keywords: Alzheimer’s disease · MRI · Convolutional neural network · Explainable boosting machine · Interpretable AI. Code availability: To be added to: [https://gitlab.com/radiology/neuro/wenjie-project](https://gitlab.com/radiology/neuro/wenjie-project) 1 Introduction Dementia is a major global health problem [21] . However, early and accurate diagnosis of AD (Alzheimer’s Disease) is challenging [24] . Machine learning methods have shown large potential for early detection and prediction of AD because they can learn subtle patterns and capture slight tissue alterations in highdimensional imaging data [7,25] . Nevertheless, those machine learning methods with high diagnostic performance, such as deep learning, are considered black boxes because of the poor interpretability of the predicted results [3] . On the other hand, intrinsically interpretable methods can provide explainable results but often have worse predictive performance as they cannot fully exploit the high-dimensional data [2] . To this end, to facilitate the translation of machine learning to clinical practice it is crucial to find an optimal tradeoff between the accuracy and

FAQs

Why are many imaging-based machine learning models for Alzheimer’s diagnosis considered black boxes?

Because it is often unclear how they make decisions from imaging inputs, leading to poor interpretability of the predicted results.

What does the proposed framework combine to improve both interpretability and accuracy?

It combines Explainable Boosting Machines with deep learning-based feature extraction to produce deep learning imaging biomarkers, then uses EBM to provide feature importance for interpretability.

How was the proposed method validated and how did it perform?

It was validated on the ADNI dataset for AD vs control classification and also tested on an external dataset for AD vs SCD classification, achieving high accuracy and AUC in both settings, and outperforming baselines using volume biomarkers and an optimized CNN comparison.

Aug 3, 2026

Upload by Patrick