Research Lab

Toward the plausible mechanism underlying rehabilitation-induced recovery after chronic ischemic stroke

Collaborators: Prof. Kenneth Fong (Department of Rehabilitation Sciences, PolyU), and Prof. Chetwyn Chan (Department of Psychology, HKEdU).

The societal burden of stroke patients with persistent neurological deficit is high. It is imperative that the mechanisms of rehabilitation-induced motor recovery be better understood in a hope to develop more efficacious rehabilitative therapy.

Treatment outcomes of people with stroke after rehabilitation vary, with up to 60% of people having residual impairment of upper limb function. The high variability in rehabilitation-induced recovery prompted researchers and clinicians to develop more efficacious rehabilitative interventions for functional regain in post-stroke patients. However, the mechanisms underlying post-stroke functional regain have not been well articulated.

Given the gap in the elusive neural processes and mechanisms underlying rehabilitation-induced recovery, we aim to gain better understanding on the functional regain of post-stroke patients by constructing a brain recovery model. As a first attempt, we propose to build the basic recovery model based on patients who will undergo constraint-induced movement therapy, a popular evidence-based post-stroke intervention, for capturing the training-induced neuroplastic changes. The two objectives of this research are: (1) to characterize the longitudinal changes in functional and structural brain networks which would differentiate the rate of changes in these networks; and (2) to define the coupling between functional and structural brain connections, and their contributions to the daily function regain.

Generation of annotated radiology report of chest X-ray using artificial intelligence

Collaborators: Prof. Tiffany Yuen-Tung So (Department of Imaging and Interventional Radiology, CUHK), Prof. Qi Duo (Department of Computer Science and Engineering, CUHK), and Prof. Xiaofan Zhang (Shanghai Jiao Tong University, and Shanghai AI Laboratory).

The aging Hong Kong population will significantly burden the public health care system in the next decade. Prime scenario of which is the rising demand on chest radiography, the most common imaging examination for the screening and management of cardiothoracic conditions. Due to the high volume of chest X-rays, the turnaround time for formal radiology reports written by trained radiologists has been and will continue to be severely compromised.

In 2019, there were three cases of oversight of abnormal X-ray findings that prompted a formal investigation by the Hospital Authority. The Investigation Panel concluded that the root cause pertained to perceptual error, and recommended the new means to support front-line doctors for timely interpretation of chest X-rays. In particular, the panel recommended to explore the use of artificial intelligence (AI) for lesion detection.

Toward these ends, there is an urgent need to develop technology to improve the throughput and quality of radiology reporting in the hope of improving disease diagnosis and monitoring of disease progression, and minimizing medical mishappening amidst the ever increasing strain on trained radiologists. The overarching goal of this research is therefore to fill the critical gaps in the radiology reporting of chest X-rays by developing a robust and novel radiology reporting AI to support clinical decision making and radiological training.

Chest X-ray artificial intelligence assistance for radiologist (10% completion)

Collaborators: Prof. Tiffany Yuen-Tung So (Department of Imaging and Interventional Radiology, CUHK), Dr. Qihua Yang (Department of Radiology, Sun Yat-sen University), and Dr. Keith CW Chiu (Department of Diagnostic and Interventional Radiology, Queen Elizabeth Hospital).

Aging Hong Kong population will exacerbate the burden on the public healthcare system, such as a rising demand on chest X-ray (CXR) examination. Due to the high volume of CXR and the shortage of radiologists, the Hospital Authority, a local statutory body which manages all public hospitals of Hong Kong, does not mandate CXR radiology report. Unreported exam predispose emergency departments to critical challenges as clinical management of undifferentiated patient is contingent on the critical radiological findings on CXR. A plausible outcome pertains to altering treatment and callbacks of patients discharged from emergency department should there be clinically significant discrepancy between emergency department physicians and radiologists.

The overall goal of this research is to develop artificial intelligence (AI) for CXR interpretation to partially fill the role of radiologist to avoid delay to critical care and compromise to therapeutic efficacy in the emergency settings. Our AI assistant can perform multiple tasks for the interpretation of key actionable CXR findings, namely cardiomegaly, edema, effusion, infiltrate, pneumothorax and devices (3). It is an interactive AI assistant designed to not only seamlessly integrate in the clinical workflow of emergency radiology, but also to improve itself by accounting for user feedback via reinforcement learning. It will be a potential alternative to assist emergency department physicians and radiologists for interpreting CXR.

1. Wang Y, Zeng T, Liu F, Dou Q, Cao P, Chang HC, Deng Q, Hui ES∗. Illuminating the Unseen: Advancing MRI Domain Generalization Through Causality. Medical Image Analysis. 2025 Jan 16:103459. (Impact factor: 10.7; Q1 journal in Radiology, nuclear medicine & medical imaging).
2. Wang Z, Lee K, Deng Q, So TY, Chiu WH, Zhou B, Hui ES*. Expert Insight-Enhanced Follow-Up Chest X-ray Summary Generation. International Conference on Artificial Intelligence in Medicine. Cham: SpringerNature Switzerland, 2024; 181-193.
3. Zhang H, Cao P, Mak HKF, Hui ES*. The structural-functional-connectivity coupling of the aging brain. Geroscience. 2024 Aug;46(4):3875-3887. (Impact factor: 5.6; Q2 journal in geriatrics & gerontology).
4. Hui ES*. Advanced diffusion MRI for prediction of stroke recovery. J Magn Reson Imaging. J Magn Reson Imaging. 2023 May;57(5):1312-1319. (Invited Review; Impact factor: 4.4; Q1 journal in radiology, nuclear medicine & medical imaging)
5. Li T, Cui D, Hui ES*, Cai J. Time-resolved magnetic resonance fingerprinting for radiotherapy motion management. Med Phys. 2020 Dec;47(12):6286-6293. (Impact factor: 3.8; Q2 journal in radiology, nuclear medicine & medical imaging)