AD-HOLDER: Alzheimer’s Disease Detection via Machine Learning based Histograms of Light GBM Classifier using MRI and PET Images

S Mahalakshmi; K Valarmathi

doi:10.2478/msr-2025-0043

.blurhash-client-img { display: none !important; }

AD-HOLDER: Alzheimer’s Disease Detection via Machine Learning based Histograms of Light GBM Classifier using MRI and PET Images

Measurement Science Review

Volume 25 (2025): Issue 6 (December 2025)

By: S Mahalakshmi and K Valarmathi

Open Access

|Dec 2025

Figures & Tables

Architecture diagram of the HOG-based LGBM model.

Table 1.

Hyperparameter settings of the LGBM classifier.

Hyper parameter	Value
Number of estimators	500
Learning rate	0.05
Maximum depth	8
Feature fraction	0.8
λ_L1(L1 reg)	0.1
λ_L2(L2 reg)	0.2
Cross-validation folds	3 and 5

Experimental results of the proposed AD-HOLDER methodology.

Performance analysis for two-class classification.

Table 2.

Efficiency assessment of the proposed AD-HOLDER model.

Classes	ACC	SPE	PRE	REC	F1
Normal	98.96	98.42	96.83	89.73	95.38
Abnormal	99.29	97.17	98.32	95.48	97.72

*ACC* curve for the proposed AD-HOLDER model.

Loss curve for the proposed AD-HOLDER model.

ROC curve of the proposed AD-HOLDER classification model.

Table 3.

Cross-validation results of the proposed AD-HOLDER model.

Metric	3-fold cross-validation [%]	5-fold cross-validation [%]
ACC	98.27	98.59
SPE	96.80	96.14
PRE	96.17	96.73
REC	89.27	90.83
F1	94.68	94.91

Table 4.

Quantitative comparison of denoising methods on MRI and PET images.

Method	PSNR [dB]	SSIM
Median filter	27.42	0.712
Gaussian filter	28.06	0.815
Bilateral filter	28.75	0.822
NLM filter	29.74	0.858
DIP (proposed)	31.82	0.917

Table 5.

Comparative analysis between traditional ML networks.

Techniques	ACC	SPE	PRE	REC	F1
KNN	90.63	89.39	90.37	85.92	87.93
CNN	93.95	94.21	92.49	87.37	90.62
Naive Bayes	92.84	91.83	88.93	86.75	89.51
Decision Tree	93.92	93.72	94.93	88.23	92.73
ViT	96.38	92.63	94.26	90.18	95.29
RF	95.29	95.90	95.27	91.35	94.62
LGBM	99.12	97.79	97.57	92.60	96.55

Table 6.

Comparison of traditional segmentation algorithms [%].

Methods	ACC	DI	IoU
U-net	91.37	71.82	58.2
V-net	93.75	81.46	69.9
Nested V-net	95.01	84.28	72.9
SegNet	92.37	86.10	73.5
GBS (ours)	99.12	90.74	82.7

Comparison results of different segmentation techniques.

Table 7.

Comparison of existing models and the proposed AD-HOLDER model.

Authors	Methods	ACC [%]	PRE [%]	REC [%]	F1 [%]	p-value
Battineni, G. [14]	GBA	97.58	94.89	88.26	90.36	0.041
Odusami, M., et al. [16]	XAI	73.90	92.67	85.29	88.82	0.37
Hamdi, M. [19]	CNN	96.00	95.92	90.51	93.52	0.42
Proposed model	AD-HOLDER model	99.12	97.57	92.60	96.55	0.029

Real-time clinical setting of the proposed AD-HOLDER model.

References

Authors

Metrics

Articles in this issue

DOI: https://doi.org/10.2478/msr-2025-0043 | Journal eISSN: 1335-8871

Journal RSS Feed

Language: English

Page range: 389 - 399

Submitted on: Oct 1, 2024

Accepted on: Sep 29, 2025

Published on: Dec 23, 2025

Published by: Slovak Academy of Sciences, Institute of Measurement Science

In partnership with: Paradigm Publishing Services

Publication frequency: Volume open

Keywords:

MRI images,

PET images,

Histogram of Oriented Gradients,

machine learning,

Light Gradient Boosting Machine,

graph-based segmentation

Related subjects:

Engineering,

Electrical engineering,

Control engineering, metrology and testing

© 2025 S Mahalakshmi, K Valarmathi, published by Slovak Academy of Sciences, Institute of Measurement Science
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.

Volume 25 (2025): Issue 6 (December 2025)

AD-HOLDER: Alzheimer’s Disease Detection via Machine Learning based Histograms of Light GBM Classifier using MRI and PET Images

Figures & Tables

Fig. 1.

Fig. 2.

Table 1.

Fig. 3.

Fig. 4.

Fig. 5.

Table 2.

Fig. 6.

Fig. 7.

Fig. 8.

Fig. 9.

Table 3.

Table 4.

Table 5.

Table 6.

Fig. 10.

Table 7.

Fig. 11.

Paradigm

My account