30 A research team led by Professor LI Hai from Hefei Institutes of Physical Science, Chinese Academy of Sciences, has developed an explainable Artificial Intelligence (AI) approach to assess the risk of central lymph node metastasis in patients with Papillary Thyroid Carcinoma (PTC) before surgery. The study was published in the Journal of Imaging Informatics in Medicine. PTC is the most common type of thyroid cancer and generally has a favourable prognosis. However, it can spread to nearby lymph nodes, making early assessment important for treatment planning. Central lymph nodes can be difficult to detect with conventional ultrasound, particularly when they are small or located in challenging anatomical positions. To improve preoperative prediction, the researchers combined information from different ultrasound techniques with clinical data. The team analysed ultrasound images and clinical information from 428 patients with 508 PTC nodules at four hospitals. The resulting machine-learning model performed well in an independent external test, achieving an AUC of 0.844. Further analysis showed AUCs of 0.975 in the training set and 0.917 in the validation set. The multimodal model outperformed single-modality and simplified fusion approaches, while external validation across multiple hospitals strengthens confidence in its generalisability. Unlike models that provide only a risk score, the new approach also identifies the imaging features that contribute to each prediction, allowing radiologists to review the evidence behind the AI assessment. Model interpretability was achieved using SHapley Additive exPlanations (SHAP), which provides feature-level explanations to support clinical interpretation. SHAP analysis identified age, texture-derived radiomic features, and elastography-derived stiffness-related features as key contributors to the predictions. This focus on transparency addresses a common limitation of medical AI systems, often described as black-box models that provide predictions without revealing their reasoning. By making the supporting imaging evidence visible, the new system allows clinicians to scrutinise and integrate the AI output with their own expertise. “The value of medical AI lies not only in predicting risk, but also in making the evidence behind each prediction visible and open to clinical scrutiny,” said Professor Hai. “By connecting multimodal imaging with radiologists’ expertise, explainable AI can support more transparent and collaborative clinical decision-making.” The researchers further tested the system with six radiologists with different levels of experience. In a controlled reader study, the radiologists independently evaluated cases under three conditions: without AI assistance, with basic AI assistance providing only a probability output, and with explainable AI assistance that included visualised feature-level contributions. With explainable AI support, the radiologists achieved higher diagnostic accuracy and showed greater agreement in their assessments. The explainable version also increased diagnostic confidence and raised human-AI agreement to substantial or almost-perfect levels across experience groups. The findings highlight the potential of explainable AI to enhance rather than replace clinical judgement. Radiologists with varying levels of experience benefited from the system, suggesting it could help standardise assessments and reduce variability in preoperative decision-making. As AI tools continue to enter clinical workflows, approaches that prioritise interpretability may gain greater acceptance among healthcare professionals.