Qinkai Yu

Qinkai Yu

Auto-generated profile, based on publicly available information

Qinkai Yu

Division of Biomedical Sciences, Bioengineering Program, King Abdullah University of Science and Technology (KAUST), Thuwal, Saudi Arabia

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
28 January 2026 in IEEE Transactions on Image Processing
Verified
Journal Article
Closed access
QYQinkai Yu
Qinkai Yu
CZChong Zhang
Chong Zhang
GJGaojie Jin
Gaojie Jin
THTianjin Huang
Tianjin Huang
WZWei Zhou
Wei Zhou
YMYanda Meng
Yanda Meng
Annotating medical data for training AI models is often costly and limited due to the shortage of specialists with relevant clinical expertise. This challenge is further compounded by privacy and ethical concerns associated with sensitive patient information. As a result, well-trained medical segmentation models on private datasets constitute valuable intellectual property requiring robust protection mechanisms. Existing model protection techniques primarily focus on classification and generative tasks, while segmentation models—crucial to medical image analysis— remain largely underexplored. In this paper, we propose a novel, stealthy, and harmless method, StealthMark, for verifying the ownership of medical segmentation models under black-box conditions. Our approach subtly modulates model uncertainty without altering the final segmentation outputs, thereby preserving the model's performance. To enable ownership verification, we incorporate model-agnostic explanation methods, e.g. LIME, to extract feature attributions from the model outputs. Under specific triggering conditions, these explanations reveal a distinct and verifiable watermark. We further design the watermark as a QR code to facilitate robust and recognizable ownership claims. We conducted extensive experiments across four medical imaging datasets (CMR dataset from UK Biobank, the SEG fundus dataset, the EchoNet echocardiography dataset, and the PraNet colonoscopy dataset) and five mainstream segmentation models. The results demonstrate the effectiveness, stealthiness, and harmlessness of our method on the original model's segmentation performance. For example, when applied to the SAM model, Stealth-Mark consistently achieved attack success rates (ASR) above 95% across various datasets while maintaining less than a 1% drop in Dice and AUC scores—significantly outperforming backdoor-based watermarking methods and highlighting its strong potential for practical deployment. Our implementation code is made available at https://github.com/Qinkaiyu/StealthMark.

10 November 2025
Verified
Conference Paper
Open access
QZQingcheng Zeng
Qingcheng Zeng
MJMingyu Jin
Mingyu Jin
QYQinkai Yu
Qinkai Yu
ZWZhenting Wang
Zhenting Wang
WHWenyue Hua
Wenyue Hua
Yongfeng ZhangYongfeng Zhang
Yongfeng Zhang

12 January 2026 in The Lancet Regional Health - Southeast Asia
Verified
Journal Article
Open access
QYQinkai Yu
Qinkai Yu
JAJinbert L. Azariah
Jinbert L. Azariah
ZAZ. Sajan Ahmad
Z. Sajan Ahmad
RARajappan Anilkumar
Rajappan Anilkumar
PCPeter Calvert
Peter Calvert
Gregory LipGregory Lip
Gregory Lip
Stroke risk stratification in patients with atrial fibrillation (AF) is challenging, particularly in under-represented South Asian populations. The use of a multimodal deep-learning artificial intelligence (AI) model, which integrates clinical data with widely available paper electrocardiogram (ECG) images, represents a novel predictive approach that has not previously been validated in this population. This study used data from the prospective KERALA-AF registry, the largest prospective AF study in South Asia. We developed a multimodal deep-learning AI model to predict incident stroke within one year by combining tabular clinical data with scanned paper ECGs. We benchmarked its performance (AUC) against machine learning (ML) models using only clinical data and the CHA2DS2-VASc score. Of 631 patients included (mean age 64.4, SD 12.9; 54.2% female), 25 (4.0%) experienced a stroke within one year. The multimodal deep learning AI model incorporating ECG data achieved the highest discrimination (AUC 0.816, 95% CI 0.704–0.914), substantially outperforming the CHA2DS2-VASc score (AUC 0.666) and all compared machine learning models trained on clinical data alone. Permutation analysis showed the scanned paper ECG images contributed 57.1% of the model's predictive signal, which boosted the model's performance significantly. Integrating scanned paper ECGs with clinical data via deep learning methods significantly enhanced 1-year stroke clinical risk prediction in South Asian AF patients. This study demonstrates the value of using multimodal AI with readily available, non-digital data in improving clinical risk stratification beyond current approaches based on clinical risk factors alone. Kerala Chapter of Cardiological Society of India.

29 May 2026 in Information Fusion
Verified
Journal Article
Closed access
JLJing Li
Jing Li
HDHongyuan Diao
Hongyuan Diao
QYQinkai Yu
Qinkai Yu
FZFeixiang Zhou
Feixiang Zhou
XZXiaorou Zheng
Xiaorou Zheng
SDShoubin Dong
Shoubin Dong

23 September 2026 in Medical Image Analysis
Verified
Journal Article
Closed access
JLJing Li
Jing Li
QYQinkai Yu
Qinkai Yu
FZfeixiang zhou
feixiang zhou
DZDongbo Zhang
Dongbo Zhang
Bo HuangBo Huang
Bo Huang
YMYanda Meng
Yanda Meng

9 April 2026 in JACC: Asia
Verified
Journal Article
Open access
QYQinkai Yu
Qinkai Yu
YMYanda Meng
Yanda Meng
Gregory LipGregory Lip
Gregory Lip