[Paper Review] iPAL: A Machine Learning Based Smart Healthcare Framework For Automatic Diagnosis Of Attention Deficit/Hyperactivity Disorder (ADHD)

Abhishek Sharma, Arpit Jain+4 more|arXiv (Cornell University)|Feb 1, 2023

EEG and Brain-Computer InterfacesNeuroscience3 citations

TL;DR

This paper proposes iPAL, a novel ensemble machine learning framework that combines phenotypic data (using SVM) and resting-state fMRI data (using a neural network) to enable accurate, automated ADHD diagnosis. The system achieves 92.5% accuracy on 400 subjects from the ADHD200 dataset, with a 99.16% accuracy from phenotypic SVM and 86.67% from fMRI-based neural networks, demonstrating strong potential for remote, scalable clinical diagnosis.

ABSTRACT

ADHD is a prevalent disorder among the younger population. Standard evaluation techniques currently use evaluation forms, interviews with the patient, and more. However, its symptoms are similar to those of many other disorders like depression, conduct disorder, and oppositional defiant disorder, and these current diagnosis techniques are not very effective. Thus, a sophisticated computing model holds the potential to provide a promising diagnosis solution to this problem. This work attempts to explore methods to diagnose ADHD using combinations of multiple established machine learning techniques like neural networks and SVM models on the ADHD200 dataset and explore the field of neuroscience. In this work, multiclass classification is performed on phenotypic data using an SVM model. The better results have been analyzed on the phenotypic data compared to other supervised learning techniques like Logistic regression, KNN, AdaBoost, etc. In addition, neural networks have been implemented on functional connectivity from the MRI data of a sample of 40 subjects provided to achieve high accuracy without prior knowledge of neuroscience. It is combined with the phenotypic classifier using the ensemble technique to get a binary classifier. It is further trained and tested on 400 out of 824 subjects from the ADHD200 data set and achieved an accuracy of 92.5% for binary classification The training and testing accuracy has been achieved upto 99% using ensemble classifier.

Motivation & Objective

• To develop a scalable, automated diagnostic system for ADHD that reduces reliance on clinician-dependent evaluations.
• To improve diagnostic accuracy by integrating multimodal data—phenotypic information and functional MRI (fMRI) scans.
• To address the limitations of traditional ADHD diagnosis, which often misdiagnoses symptoms due to overlap with other neurodevelopmental disorders.
• To enable remote and real-time diagnosis through a cloud-hosted, intelligent healthcare framework suitable for underserved populations.
• To validate the framework on a large, publicly available dataset (ADHD200) while ensuring robustness through data preprocessing and feature engineering.

Proposed method

• Utilized the ADHD200 dataset, including phenotypic data from 505 subjects and fMRI data from 400 subjects after data curation.
• Applied Support Vector Machine (SVM) with radial basis function (RBF) kernel for multiclass and binary classification on phenotypic data.
• Trained a deep neural network on functional connectivity matrices derived from resting-state fMRI (rs-fMRI) data using the nilearn library and later scaled to 400 subjects via custom cloud-hosted pipeline.
• Employed feature selection techniques including meanFD and DVARS for fMRI, and RMSFD for motion correction, to reduce noise and improve model generalization.
• Combined the two classifiers (SVM on phenotypic data and ANN on fMRI) using an ensemble averaging technique with equal weighting for final binary classification.
• Validated models using 10-fold cross-validation and reported performance metrics including accuracy, precision, and F1-score on test splits.

Experimental results

Research questions

• RQ1 Can a machine learning ensemble combining phenotypic and fMRI data significantly improve ADHD diagnosis accuracy compared to single-modality approaches?
• RQ2 How does the performance of SVM on phenotypic data compare to other supervised learning models (e.g., logistic regression, KNN, AdaBoost) in classifying ADHD?
• RQ3 To what extent can a deep neural network trained on functional connectivity from rs-fMRI data achieve high diagnostic accuracy without prior neuroscience knowledge?
• RQ4 Does integrating phenotypic and neuroimaging data via ensemble learning lead to more robust and generalizable diagnosis than using either modality alone?
• RQ5 Can the framework be scaled to larger datasets while maintaining high performance, and what are the challenges in data standardization and noise reduction?

Key findings

• The SVM classifier on phenotypic data achieved 99.16% accuracy, outperforming logistic regression, KNN, AdaBoost, and Random Forest on the same dataset.
• The deep neural network on fMRI data achieved 86.67% accuracy on the 400-subject fMRI subset, demonstrating strong performance despite limited initial data (n=40) from nilearn.
• The ensemble model combining phenotypic SVM and fMRI neural network achieved 92.5% accuracy with equal weighting, showing synergistic improvement over individual models.
• The ensemble model achieved 99% training accuracy and 92.5% test accuracy, indicating strong generalization and low overfitting.
• The framework achieved 100% precision and 99% F1-score on the ensemble model, confirming high reliability in both positive and negative predictions.
• The study confirms that phenotypic data alone provides a highly accurate diagnostic signal, while fMRI data adds complementary predictive power when properly preprocessed and integrated.