Multimodal AI Diagnostic Decision Support Systems

Google and MayoClinic are listed as corporate participants, while Linus Health Inc. is a named startup; research contributors include Shanghai JiaoTong University and Zhejiang University. Shanghai JiaoTong University has 55 recent-window documents, Zhejiang University 49, and Google 54. These are activity signals, not proof that each organization has a deployed clinical product.

FACTS identifies 8 startups, but names only AILab and Linus Health Inc. in its startup profiles, so it does not support a reliable ranking of the most promising firms for 2026. Linus Health has 2 patents in the window and works on cognitive and movement-disorder decision support. AILab’s work includes ECG question-answering benchmarks and explainable support for chronic eye disease.

Shanghai JiaoTong University and Zhejiang University are among the most active named research organizations, with 55 and 49 recent-window documents respectively. Other active groups include Tsinghua University, which works on mammography vision-language models, and the Chinese Academy of Sciences CAS, whose research includes retrieval-augmented radiology report generation. These document counts indicate activity, not clinical adoption.

The recorded funding data lists 15 grants since 2020, totaling $38,842,094. The figure should be interpreted cautiously: the funding feed includes mixed-purpose awards and does not establish that every dollar was dedicated to multimodal diagnostic support. It is a recorded funding signal, not a comprehensive measure of investment across the field.

OneSource Solutions International Inc. and Linus Health Inc. are the leading listed patent holders, with 2 patents each. Other named holders include CAS Shenzhen Institutes of Advanced Technology, Clover Health, Curai Inc., Rad Ai, O.vision and Persown Inc., each with 1 listed patent. The patent-holder list is separate from the field’s 14 patents recorded in the 24-month window.

No: the 14 patents recorded in the 24-month window accompany a 25% decline in patent activity over the latest 12 months versus the previous 12 months. By contrast, publications reached 1,565 in the last year and grew 340.8% over the same comparison window. The difference points to a research-heavy field, not proof of commercial readiness.

In September 2026, SR University researchers proposed a quantum–deep hybrid architecture for processing healthcare signals and images, while Adnan Aydin University researchers explored explainable Alzheimer’s and frontotemporal dementia diagnosis using EEG and clinical data. Other recent work includes JADE-Plus, a multimodal agentic retrieval-augmented framework for jawbone lesion diagnostic support, published in July 2026.

The 5 listed approaches are early-level fusion, joint representation learning, late-level fusion, vision-language models and knowledge-grounded reasoning. Early fusion combines inputs up front; joint learning builds shared representations; late fusion combines separate predictions. Vision-language models link images and text, while knowledge-grounded systems retrieve clinical evidence to support reasoning. Each approach trades flexibility or richer interactions against data, alignment or grounding challenges.

The listed applications include neurology, oncology, radiology, primary and emergency care, and digital pathology. Neurology has 690 documents, oncology 578, radiology 505, primary and emergency care 458, and digital pathology 127. Recent examples span EEG-based dementia assessment, breast cancer diagnosis using mammography and pathology images, and multimodal decision support for clinical care.