The June Affinity Group hosts Tianyu Zhang, PhD — postdoctoral researcher at the Netherlands Cancer Institute (Amsterdam) and Radboud University Medical Centre (Nijmegen). His talk presents LUMIN: A Longitudinal Multi-modal Knowledge Decomposition Network for Predicting Breast Cancer Recurrence, recently presented at AAAI 2026.

LUMIN is one of the first frameworks to treat post-treatment recurrence as a downstream prediction task built directly on routine surveillance data — longitudinal mammograms paired with electronic health records. Two ideas drive its accuracy. A Cross-Modal Disentangled Knowledge Extractor separates shared, complementary, and modality-specific signal across imaging and text rather than collapsing them with naïve fusion. A Temporal Evolution Disentangled Knowledge Extractor splits time-invariant, time-varying, and time-specific features so the model can reason about disease dynamics rather than a single static snapshot.

Trained on 3,924 patients and 19,684 exams from the Netherlands Cancer Institute, LUMIN reaches an AUC of 0.729 and stratifies patients into risk groups whose 5-year recurrence rates differ by 3.4× (HR 3.21, 95% CI 2.13–4.84). Beyond the metric gains, the talk is methodologically interesting for the breast-cancer risk-AI community: it defines recurrence prediction as a tractable task on screening data, and it shows how multi-step, multimodal inputs can be decomposed into clinically interpretable factors.

The talk is Friday, June 5, 2026 at 9:00 AM Hawaiʻi time, on Zoom.

Register for this Affinity Group meeting →

Register once and Zoom will email you the join link plus reminders for this and any future Affinity Group meeting. Recordings are posted to the AI PHI YouTube channel ↗. To suggest a future speaker, email aiphi@shepherdresearchlab.org.