Explainable Artificial Intelligence for Human-Machine Interaction in Brain Tumor Localization

Area of Science:

• Neuro-oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Primary brain malignancies are a leading cause of death globally.
• Artificial intelligence (AI) and computer vision offer advanced tools for brain tumor characterization and diagnosis.
• Current AI models often function as black boxes, hindering clinical adoption due to a lack of interpretability.

Purpose of the Study:

• To evaluate deep learning algorithms for localizing brain tumor lesions in MRI contrasts.
• To assess the ability of AI to distinguish tumorous regions from healthy brain tissue.
• To investigate the role of explainable AI in improving model interpretability and performance evaluation.

Main Methods:

• Selected deep learning algorithms were applied to magnetic resonance imaging (MRI) data.
• The study focused on lesion localization and classification accuracy.
• Correlation analysis was performed between classification and lesion localization.

Main Results:

• A significant correlation (R = 0.46, p = 0.005) was observed between classification and lesion localization accuracy.
• Some AI algorithms incorrectly classified tumors based on non-relevant features.
• The findings highlight limitations in current AI interpretability for clinical application.

Conclusions:

• Explainable AI approaches are essential for building trust and understanding AI model decisions in clinical settings.
• Developing interpretable AI is vital for accurate performance evaluation and optimizing training methods.
• Enhanced human-machine interaction through explainable AI can improve diagnostic accuracy and clinical workflow.