KANEx: Translating Kolmogorov-Arnold Networks' Interpretability to Medical Explainability

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Krithi Shailya, Ananya Lakshmi Ravi, Venkatanathan K. V., Sowmya S. Sundaram, Gokul S. Krishnan, Aditi Anand, Balaraman Ravindran
• Venue: arXiv: cs.CV
• Published: 2026-07-27
• Collected: 2026-07-28
• URL: http://arxiv.org/abs/2607.24730v1
• PDF: https://arxiv.org/pdf/2607.24730v1
• DOI: UNKNOWN

Plain-Language Brief

A new framework, KANEx, is introduced to leverage the inherently interpretable components of Kolmogorov-Arnold Networks (KANs) for generating more reliable textual explanations in chest X-ray classifiers. The approach improves semantic similarity and visual localization by 10%, potentially increasing trust in medical AI.

What the Document Says

• The paper discusses the use of a new framework named KANEx, which uses Kolmogorov-Arnold Networks (KANs) to generate more reliable textual explanations for chest X-ray classifiers.
• KANEx is designed to ground Visual Language Models reasoning using the symbolic transparency of KANs.
• A new heatmap generation method, KAN-Map, is introduced, derived directly from KAN models rather than gradient approximations.
• The grounded contexts are fed into downstream VLMs for enhanced explainability.
• Benchmarking on the MIMIC-CXR dataset, the paper shows that KAN-based architectures with ResNet/ViT baselines demonstrate improved semantic similarity and produce significantly more faithful saliency maps.
• The study finds that KAN architectures improve visual localization and downstream reasoning quality by 10%.

Key Concepts Mentioned

• Kolmogorov-Arnold Networks (KANs)
• KANEx
• Vision-Language Models (VLMs)
• chest X-ray classifiers
• MIMIC-CXR dataset
• semantic similarity
• saliency maps
• heatmap generation

Claims Made by the Document

• KANEx can generate more reliable textual explanations for chest X-ray classifiers.
• KAN architectures improve visual localization and downstream reasoning quality by 10%.

Evidence Used

• Improved semantic similarity and faithful saliency maps demonstrated on the MIMIC-CXR dataset

Methods / Approach

• Use of KANEx framework to ground VLM reasoning
• Introduction of KAN-Map for heatmap generation

Limitations Stated by the Authors

Unknown

Limitations Noted by the Agent

Agent interpretation — not from the source.

It is unclear if the paper discusses potential limitations of the proposed method.

Open Questions

• What other medical applications could benefit from this approach?
• Are there any potential drawbacks or risks associated with using KAN-based architectures in healthcare?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The paper focuses on chest X-ray classifiers, but it is possible that the method could be applicable to other medical imaging tasks.

Raw Abstract / Excerpt

Computer vision models have become highly effective for medical applications, yet their black-box nature continues to undermine clinician trust. In clinical workflows, chest X-ray classifiers are increasingly paired with Vision-Language Models (VLMs) to generate natural-language explanations. However, these systems add linguistic fluency without addressing the underlying opacity of the visual model. With the emergence of Kolmogorov-Arnold Networks (KANs), whose spline-based components provide inherently interpretable functional units, we investigate whether this architectural transparency can be leveraged to produce more trustworthy textual explanations. We introduce KANEx, the first ever framework that leverages the symbolic transparency of KANs to ground VLM reasoning. This interpretability also made it possible to design KAN-Map, a novel heatmap generation method derived directly from KAN models rather than gradient approximations. We feed these grounded contexts into downstream VLMs for enhanced explainability. Benchmarked on the MIMIC-CXR dataset, we demonstrate that KAN-based architectures with ResNet/ViT baselines demonstrate improved semantic similarity while producing significantly more faithful saliency maps. KAN architectures improve visual localization and downstream reasoning quality by 10%. Our findings suggest that grounding linguistic explanations and visual attributions in mathematically interpretable units is a necessary step toward trustworthy medical AI.