نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

Explainable Artificial Intelligence in Brain Mapping: Opportunities, Challenges, and Future Directions

Subject:

Artificial Intelligence and Brain Mapping

Presenter Author:
Name:	Mohamadreza Nouraie

Author:
Name 1:	Mohamadreza Nouraie - Corresponding - Presenter
Affiliation 1:	Student Research Committee, Amol Faculty of Allied Medical Sciences, Mazandaran University of Medical Sciences, Sari, Iran.

Name 2:	Zahra Arabkermani
Affiliation 2:	Department of Paramedicine, Amol School of Paramedical Sciences, Mazandaran University of Medical Sciences, Sari, Iran

Name 3:
Affiliation 3:

Name 4:
Affiliation 4:

Name 5:
Affiliation 5:

Name 6:
Affiliation 6:

Name 7:
Affiliation 7:

Name 8:
Affiliation 8:

Name 9:
Affiliation 9:

Name 10:
Affiliation 10:

Abstract

Original Research

introduction :

Explainable Artificial Intelligence (XAI) has been regarded as an important approach in neuroimaging and brain mapping research. Furthermore, it has answered the question about the reasons for AI-based clinical decisions. Even though deep learning and machine learning algorithms have great potential to predict neurological disorders, such as Alzheimer’s disease, brain tumors, and strokes, their “black-box” nature has limited understanding and weakened clinical trust. Recent reports indicate an enthusiasm for using XAI techniques, especially saliency maps, gradient-weighted class activation mapping (Grad-CAM), local interpretable model-agnostic explanations (LIME), and SHapley Additive exPlanations (SHAP), to analyze brain imaging data, detect brain regions affected, and make more informed clinical choices. Despite promising findings, there are challenges in evaluating the quality of explanations, ensuring reproducibility within a wide range of imaging methods, and achieving a balance between predictability and explainability.

Methods:

A search was performed for peer-reviewed articles in PubMed, Scopus, and Google Scholar from 2019 to 2025. The search terms included “explainable AI” combined with terms such as “neuroimaging”, “brain mapping”, and “interpretability”. The inclusion criteria were original research using human brain imaging data (MRI, fMRI, EEG, PET, DTI), which reported prediction performance (accuracy, AUC, sensitivity, specificity) and explainability output, and were published in English. Articles which presented only methodological approaches, had no imaging data or did not assess the quality of explanations, were excluded from the study. Extracted data included the type of imaging, models used, dataset size, XAI technique applied, and validation methods.

Results:

Among around 2,500 initial records, 30 studies met the inclusion criteria. Structural MRI and fMRI had the highest contribution, but the use of EEG and multimodal imaging was less frequent, with a growing trend. The predictive accuracy ranged from 60% to 98%, and AUC (Area Under the Curve) values were from 70% to 98%. The best performance was usually achieved using deep learning algorithms on large MRI datasets. Approaches based on saliency maps (Grad-CAM, CAM), feature importance measures (SHAP, permutation importance), and LIME (Local Interpretable Model-agnostic Explanations) were the most frequently used explanation tools. Explanations often emphasized important clinical regions, such as the hippocampus in Alzheimer’s disease, or tumor margins in oncology applications. However, only about a quarter of the studies matched explanatory maps with expert assessments or reference brain atlases, and reproducibility among diverse datasets remained limited.

Conclusion:

Explainable artificial intelligence presents a valuable perspective towards an increase in transparency and clinical acceptance of AI models. Main gaps include the absence of standardized criteria for evaluating explanations, limited expert verification, and difficulty in adjusting accuracy with explainability. These research models are recommended for future research: integrating multimodal imaging in heterogeneous populations, applying consensus-based criteria to assess the quality of explanations, and engaging clinical experts for human-based validation. Accepting these guidelines can pave the way for developing reliable and generalizable XAI applications in neuroimaging and promoting scientific vision and practical clinical application of brain mapping technologies.

Keywords:

Artificial Intelligence; Brain Mapping; Neuroimaging; Deep Learning; Magnetic Resonance Imagin