Decoding Neural Signatures: An Interpretable Deep Network Architecture for Detecting Attention-Deficit/Hyperactivity Disorder Using Electroencephalography

Status: In Review 2025

Rehabilitation Exercise Assessment Using Convolutional Neural Network with Feature Prioritization using Whale Optimization and Shapely Values

Status: In Review 2025

ST-EEG: A Novel Spatio-Temporal Based Artificial Intelligence Framework For Parkinson Disease Detection Using Physiological Signals

Status: In Review 2025

Rehabilitation exercise score prediction through integration of spatial feature with graph structural learning

Md. Johir Raihan

Md Golam Mustafa

Ali Saleh Alammary

Abdullah-Al Nahid

Journal Engineering Applications of Artificial Intelligence - Elsevier 2025

This research proposes a novel C-GNN architecture that fuses 2D CNNs and GNNs to automate rehabilitation exercise assessment by effectively modeling spatial and temporal dependencies. Time-series data is transformed into images using Heatmaps, MTF, and GAF, which are then divided into patches connected through RMSE-based graphs. CNNs extract local spatial features, while GNNs capture inter-patch relationships to understand movement dynamics. The model achieves strong performance with MAD scores of 0.132 on the KIMORE dataset and 0.029 on the UI-PRMD dataset

A Deep Learning Model with Interpretable Squeeze and Excitation for Automated Rehabilitation Exercise Assessment

Md. Johir Raihan

Md Atiqur Rahman Ahad

Abdullah-Al Nahid

Journal Medical & Biological Engineering & Computing - Springer 2025

The paper introduces a CNN-SE model optimized using the Grey Wolf Optimization algorithm for automated assessment of rehabilitation exercises. It demonstrates high accuracy on the KIMORE and UI-PRMD datasets across various motor tasks involving both healthy and impaired participants. To ensure transparency, SHAP is employed to interpret feature importance over time, enhancing model explainability.

Estimation of Muscle Activation during Complex Movement Using Unsupervised Motion Primitives Decomposition of Limb Kinematics

Mainul Islam Labib

Md. Johir Raihan

Abdullah-Al Nahid

Conference Activity, Behavior, and Healthcare Computing 2025

This work introduces a novel approach to decomposing complex kinematic dynamics into "motion primitives," which are then used to estimate muscle activation profiles. By analyzing the temporal evolution of these motion primitives, we offer insights into the muscle activation patterns associated with both engaged and unengaged limb movements, enhancing rehabilitation strategies for patients.

Bengali-Sign: A Machine Learning-Based Bengali Sign Language Interpretation for Deaf and Non-Verbal People

Md. Johir Raihan

Mainul Islam Labib

Abdullah Al Jaid Jim

Jun Jiat Tiang

Uzzal Biswas

Abdullah-Al Nahid

Journal Sensors - MDPI 2024

In this paper, we developed a Convolutional Neural Network combined with a Squeeze Excitation Network to accurately predict sign language signs, achieving 99.86% accuracy on the KU-BdSL dataset. Our smartphone application integrates this ML model, and we've utilized SHAP (Shapley Additive Explanations) to interpret the model's decision-making process. SHAP analysis reveals that our model focuses on hand-related visual cues, reflecting natural human communication patterns.

Detection of The Chronic Kidney Disease Using XGBoost Classifier And Explaining The Influence of The Attributes on The Model Using SHAP

Md. Johir Raihan

Md. Al-Masrur Khan

Seong-Hoon Kee

Abdullah-Al Nahid

Journal Scientific Reports - Nature 2023

Using the XGBoost classifier, we have obtained an accuracy, precision, recall, and F1 score of 99.16%, 100%, 98.68% and 99.33%, respectively using all 24 features. The BBO algorithm selected almost half of the initial features. We have obtained an accuracy, precision, recall, and F1 score of 98.33%, 100%, 97.36% and 98.67% respectively using only 13 features selected by the BBO algorithm. Finally, we have explained the impact of the feature on the ML models using the SHapley Additive exPlanations (SHAP) analysis.

Classification of Histopathological Colon Cancer Images Using Particle Swarm Optimization-Based Feature Selection Algorithm

Md. Johir Raihan

Abdullah-Al Nahid

Book Diagnostic Biomedical Signal and Image Processing Applications with Deep Learning Methods 2023

In this chapter, we present a classification approach for CC images using Particle Swarm Optimization (PSO), with a focus on maintaining low computational overhead. Additionally, we employ SHapley Additive exPlanations (SHAP) to enhance the interpretability of the machine learning models, addressing their black-box nature. Utilizing a Gradient Boosting Classifier, our approach achieved impressive performance metrics on the CC dataset: 99.73% accuracy, 99.86% precision, 99.6% recall, and a 99.73% F1-score.

Breast Cancer Classification Along With Feature Prioritization Using Machine Learning Algorithms

Abdullah-Al Nahid

Md. Johir Raihan

Abdullah Al-Mamun Bulbul

Journal Health and Technology - Springer 2022

We applied several gradient-based machine learning models—Gradient Boosting (GB), XGBoost (XGB), CatBoost (CB), and LightGBM (LGBM) to classify breast cancer and evaluate feature importance. SHapley Additive exPlanations (SHAP) were used to interpret model predictions and understand the contribution of each feature. Additionally, both filter and wrapper-based feature selection techniques were employed to prioritize relevant features. The best-performing model achieved an accuracy of 82.85%, precision of 80.00%, recall of 88.89%, and an F1-score of 84.21%.

Smart Human Following Baby Stroller Using Computer Vision

Md. Johir Raihan

Md. Tariq Hasan

Abdullah-Al Nahid

Conference International Conference on STEM and the Fourth Industrial Revolution (ICSTEM4IR) 2022

This paper proposes a human following baby carrier car that can follow a human without any kind of help from the person. This reduces the labor work for a person and provides much freedom to carry out their activities. Such a carrier can be operated by elderly people, disabled people, pregnant women, or by any other without the need for much labor. The proposed system relies on the computer vision analysis that uses the camera feed to detect the person to be followed and sends a command to a microcontroller that operates the Carrier.

Histopathological breast cancer image classification with feature prioritization using a heuristic algorithm

AA Nahid

MJ Raihan

N Sikder

SR Sabuj

Book Artificial Intelligence in Cancer Diagnosis and Prognosis, Volume 2 2022

In this work, we describe the extraction of global and local features from images using convolutional neural networks and the use of histograms to classify breast cancer images. In addition, a heuristic algorithm is used to extract a smaller number of effective features with which to classify breast cancer images using an ensemble of random forest classifiers.

Malaria Cell Image Classification by Explainable Artificial Intelligence

Md. Johir Raihan

Abdullah-Al Nahid

Journal Health and Technology - Springer 2022

In this paper, we have presented a malaria detection framework based on wavelet packet 2d, Convolutional Neural Network. (CNN), and Whale Optimization Algorithm (WOA). We have extracted a global feature set using the wavelet packet 2D and CNN. Further, we removed noisy features and selected an effective feature using WOA and XGBoost. The selected feature set is almost half of the initial feature set. Additionally, we have used SHapley Additive exPlanations (SHAP) to interpret the trained model and assess the significance of each feature. Based on the selected feature set, the XGBoost algorithm provides accuracy, precision, recall, and F1 score of 94. 78%, 94. 39%, 95. 21%, and 94. 80% respectively.

Automated Rehabilitation Exercise Assessment by Genetic Algorithm-Optimized CNN

Md. Johir Raihan

Md Atiqur Rahman Ahad

Abdullah-Al Nahid

Conference Joint 10th International Conference on Informatics, Electronics & Vision (ICIEV) and 2021 5th International Conference on Imaging, Vision & Pattern Recognition (icIVPR) 2021

In this paper, we have proposed an exercise assessment framework using the 1D Local Binary Pattern (LBP) to extract valuable features from skeleton data and a genetic algorithm (GA) optimized Convolutional Neural Network (CNN) to predict the score. The KIMORE dataset has been used in this study. We have achieved 0.0165 Mean Absolute Deviation (MAD) on the training set and 0.13515 on the validation set.