An Explainable Deep Model for Risk Scoring and Accurate Radionecrosis Identification Following Brain Metastasis Stereotactic Radiosurgery.
Authors
Affiliations (3)
Affiliations (3)
- Duke University Medical Center, Durham, NC.
- Duke Kunshan University, Jiangsu, China.
- Duke University Medical Center, Durham, NC. Electronic address: [email protected].
Abstract
As survival improves for patients with brain metastases, distinguishing local recurrence(LR) from radionecrosis(RN) is a growing neuro-oncologic challenge. We aimed to develop an explainable deep learning(DL) model to non-invasively distinguish RN from LR in patients with non-small cell lung cancer(NSCLC) following stereotactic radiosurgery(SRS). A 2<sup>nd</sup> order Heavy-Ball Neural Ordinary Differential Equation(HBNODE) DL framework was designed. It enabled dynamic tracking of input evolution within DNN, integrating MR, clinical, and genomic features into a unified Image-Genomic-Clinical(I-G-C) space. This allowed visualization of sample trajectories during model execution. Layer-Wise Relevance Propagation(LRP) was applied to quantify individual non-imaging feature contributions and their influence on diagnosis. Within the I-G-C space, a decision-making field(F) was reconstructed. The temporal evolution of F enabled quantitative comparison of cumulative contributions from each feature. Key intermediate states, defined as locoregional equilibrium points(∇F=0), were identified and aggregated using a non-parametric model to optimize prediction. High-contributing features were selected via k-means clustering of LRP results, forming a risk score model for RN vs. LR differentiation. The dataset included 142 BM lesions from 103 NSCLC patients, incorporating 3-month post-SRS T1+C MRI, seven genomic biomarkers, and seven clinical parameters. An 8:2 ratio was used for training and independent testing. Three high-contributing features, age(x1), ALK(x0.84) and PD-L1(x0.76) status, were identified by LRP and used to construct the risk score. The risk score model outperformed the model using all unweighted clinical/genomic features and an MR-only DNN. The HBNODE model, embedding the risk score within deep space, achieved the best performance across all metrics. The derived risk score, based on non-imaging features, offers a simple and rapid indicator for distinguishing RN from LR. When integrated with MRI in the HBNODE model, it further enhanced predictive performance while maintaining high explainability, highlighting its potential as a clinical decision-aid tool for BM management.