Explainable AI Framework for Automated Lesion Segmentation and Severity Assessment in Neonatal Hypoxic-Ischemic Encephalopathy

Abstract

Neonatal hypoxic-ischemic encephalopathy (HIE) is a critical brain injury in newborns resulting from perinatal oxygen deprivation, often leading to long-term neurological impairments and developmental delays. Early and accurate diagnosis is essential for improving clinical outcomes. In this work, we propose a multimodal framework for automated HIE lesion segmentation and severity classification from magnetic resonance imaging scans. A SegResNet architecture was employed to segment lesions from apparent diffusion coefficient (ADC) and Z-scored ADC maps. In the subsequent stage, an early fusion strategy was adopted to integrate features derived from the predicted lesion masks with clinical metadata. The fused representation was classified into mild, moderate, and severe HIE categories using an XGBoost classifier. The Boston Neonatal Brain Injury Dataset for Hypoxic-Ischemic Encephalopathy was used for the experiments. The severity classification received an accuracy of 88.64%, and the segmentation model received a Dice score of 0.758 ± 0.29. Explainable AI with Gradient-weighted Class Activation Mapping was also used to improve the clinical dependability. These results demonstrate the effectiveness of the proposed system in both precise lesion localization and clinically meaningful severity stratification, highlighting its potential to support early clinical decision-making in neonatal care.



Received: 12 September 2025 | Revised: 24 March 2026 | Accepted: 10 April 2026



Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.



Data Availability Statement

The data that support the findings of this study are openly available in the Zenodo repository at https://zenodo.org/records/10602767.



Author Contribution Statement

Athira Chandran: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization. Lekshmi Chandrika Reghunath: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration.