Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-1443
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-15T10:43:06.521278Z
status: published
lens: g_space
sector: health
headline: An expert routing-driven graph framework from tongue multi-task features to liver fibrosis prediction
dek: Liver fibrosis is a global health concern where early detection is crucial for timely intervention. Traditional Chinese Medicine tongue diagnosis offers a promising non-invasive assessment, but suffers from inter-observer variability and subjective interpretation. Current artificial intelligence approaches focus on binary classification without revealing tongue features influencing diagnosis, limiting clinical utility. To address these challenges, we propose two key artificial intelligence innovations: (1) Hiera…
gain_title: A tongue-image framework that identifies seven tongue attributes and models how they interact reached 0.751 accuracy and 0.767 AUC for liver fibrosis risk screening, with results better than the compared graph and multimodal models.
problem_title: (none)
trace_subject: (none)
gain_reading: A tongue-image framework that identifies seven tongue attributes and models how they interact reached 0.751 accuracy and 0.767 AUC for liver fibrosis risk screening, with results better than the compared graph and multimodal models.
gain_evidence: Our method achieves 0.751 accuracy and 0.767 area under the curve | identifies seven critical tongue attributes and models their interrelationships for liver fibrosis risk screening
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers built an AI system called Tongue Network with Dynamic Graph Relationship Learning to screen for liver fibrosis from tongue images. It combines a multi-task expert architecture with a graph neural network that builds a patient-specific relationship graph among tongue attributes.

The work matters because earlier tongue-diagnosis AI gave only binary classifications and did not show which tongue features drove a result, which limited clinical use. The visualizable outputs aim to address that gap, but the source gives only performance figures from the authors' own evaluation, so real-world clinical benefit is not established.
limitation: Accuracy of 0.751 and AUC of 0.767 are moderate, and the source reports no clinical deployment or patient outcomes, so the tool is shown only as a risk screening framework.
tag: Evidence-backed gain
key_points: The framework uses tongue images as a non-invasive input for liver fibrosis risk screening in Traditional Chinese Medicine-based hepatology practice. | It identifies seven tongue attributes and models their interrelationships through patient-specific graphs. | Reported performance is 0.751 accuracy and 0.767 area under the curve. | The authors report improvements of 0.07 and 0.08 over state-of-the-art graph neural networks and multimodal approaches. | Code is publicly available on GitHub.
rundown: The study starts from the problem that tongue diagnosis in Traditional Chinese Medicine is non-invasive but affected by inter-observer variability and subjective interpretation, while early liver fibrosis detection is important for timely intervention.

Technically, the first component is an adaptive multi-task expert network with quota routing that allocates experts by sample complexity and is described as solving negative transfer. The second converts expert selection patterns into patient-specific graphs for modeling attribute interactions.

The outputs include visualizable classifications and relationship structures, which the authors say give insight into how tongue features relate to liver fibrosis risk.
sources:
- peer_reviewed | Engineering Applications of Artificial Intelligence | https://doi.org/10.1016/j.engappai.2026.116294 | 2026-10-14
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

7198b626446309c02fe5a09c2872aa58bd30a90030a1249ca07454e1880df572

previous

0000000000000000000000000000000000000000000000000000000000000000