Heterogeneous Aligned Fusion for Survival Classification with Missing Modalities

Zheng Zheng, Yuzhi Guo, Xiao Hu, Yuwei Miao, Hehuan Ma, Jean Gao, Junzhou Huang

Keywords: Heterogeneous Aligned Fusion (HAF), multimodal learning, head and neck cancer, survival prediction, pathology imaging, MIL

TL;DR: We introduce HAF, a multimodal framework that aligns and fuses seven heterogeneous and partially missing modalities. On the HANCOCK cohort, HAF surpasses prior baselines and offers a strong benchmark for scalable clinical multimodal learning.

Abstract:

Accurate survival prediction is essential for guiding personalized treatment in head and neck cancer. Heterogeneous biomedical data, from histopathology to clinical and laboratory measurements, offer complementary prognostic value but differ in dimensionality, reside in incompatible feature spaces, and are frequently missing, making robust multimodal learning challenging.
To address this, we propose HAF (Heterogeneous Aligned Fusion), a three-stage framework for survival prediction under heterogeneous and incomplete multimodal inputs. HAF (i) uses detached prognostic supervision to obtain stable representations, (ii) performs lightweight global alignment that projects all modalities into a shared latent space while preserving patient-level discriminability, and (iii) enforces monotonic robust fusion that encourages performance to remain stable or improve when modalities are added. To the best of our knowledge, HAF is the first approach that jointly leverages all seven modalities in the HANCOCK cohort. Extensive comparisons against representative late-, early-, attention-based, and bilinear-interaction fusion methods demonstrate that HAF consistently improves both accuracy and robustness under heterogeneous and partially missing modalities.

Primary Subject Area: Integration of Imaging and Clinical Data

Secondary Subject Area: Application: Histopathology

Registration Requirement: Yes

Reproducibility: https://github.com/zz9tf/HAF

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Replace NNN with your OpenReview submission ID., Includes \documentclass{midl}, \jmlryear{2026}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package., Did not use the times package., Use the correct spelling and format, avoid Unicode characters, and use LaTeX equivalents instead., Any math in the title and abstract must be enclosed within $...$., Did not override the bibliography style defined in midl.cls and did not use \begin{thebibliography} directly to insert references., Avoid using \scalebox; use \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., No separate supplementary PDF uploads., Acknowledgements, references, and appendix must start after the main content.

Submission Number: 219