Timing Liu

Timing Liu

Auto-generated profile, based on publicly available information

Timing Liu

University of Cambridge Cambridge UK

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

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

Representative Publications

View All Publications
13 September 2023 in Nature
Verified
Journal Article
Open access
PDF file
YZYukun Zhou
Yukun Zhou
MCMark A. Chia
Mark A. Chia
SWSiegfried K. Wagner
Siegfried K. Wagner
MAMurat S. Ayhan
Murat S. Ayhan
DWDominic J. Williamson
Dominic J. Williamson
PKPearse Keane
Pearse Keane
Medical artificial intelligence (AI) offers great potential for recognizing signs of health conditions in retinal images and expediting the diagnosis of eye diseases and systemic disorders1. However, the development of AI models requires substantial annotation and models are usually task-specific with limited generalizability to different clinical applications2. Here, we present RETFound, a foundation model for retinal images that learns generalizable representations from unlabelled retinal images and provides a basis for label-efficient model adaptation in several applications. Specifically, RETFound is trained on 1.6 million unlabelled retinal images by means of self-supervised learning and then adapted to disease detection tasks with explicit labels. We show that adapted RETFound consistently outperforms several comparison models in the diagnosis and prognosis of sight-threatening eye diseases, as well as incident prediction of complex systemic disorders such as heart failure and myocardial infarction with fewer labelled data. RETFound provides a generalizable solution to improve model performance and alleviate the annotation workload of experts to enable broad clinical AI applications from retinal imaging.

19 July 2023 in Clinical Chemistry
Verified
Journal Article
Open access
PDF file
VTVivienne J Tan
Vivienne J Tan
TLTiming Liu
Timing Liu
ZAZainul Arifin
Zainul Arifin
BPBeatrice Pak
Beatrice Pak
ATArnold S. C. Tan
Arnold S. C. Tan
SCSamuel Chong
Samuel Chong
Current strategies for preimplantation genetic testing for aneuploidy or structural rearrangements (PGT-A/SR) rely mainly on next-generation sequencing (NGS) and microarray platforms, which are robust but require expensive instrumentation. We explored the suitability of third-generation single-molecule sequencing as a PGT-A/SR screening platform for both aneuploidy and segmental imbalance. Single-cell and multicell replicates from aneuploid or segmentally unbalanced cell lines (n = 208) were SurePlex-amplified, randomized, and subjected to (a) Nanopore-based single-molecule sequencing (Oxford Nanopore Technologies) and (b) NGS using a leading commercial PGT-A solution (Illumina VeriSeq PGS). Archival SurePlex-amplified trophectoderm biopsy samples (n = 96) previously analyzed using the commercial kit were blinded and reanalyzed using Nanopore. Nanopore-based PGT-A identified the specific aberration in 95.45% (84/88) and 97.78% (88/90) of single-/multicells with an aneuploidy or segmental imbalance (10–30.5 Mb), respectively. Comparison against the commercial kit’s results revealed concordances of 98.86% (87/88) and 98.89% (89/90) for the aneuploid and segmentally unbalanced (10–30.5 Mb aberration) samples, respectively. Detection sensitivity for smaller segmental imbalances (5–5.8 Mb aberration, n = 30) decreased markedly on both platforms. Nanopore-based PGT-A reanalysis of trophectoderm biopsy samples was 97.92% (94/96) concordant with the commercial kit results. Up to 24 SurePlex-amplified single-cell, multicell, or trophectoderm samples could be sequenced in a single MinION flow-cell for subsequent preimplantation genetic testing for aneuploidy or structural rearrangements (PGT-A/SR) analysis, with results obtainable in ≤3 days and at per-sample costs that are competitive with commercial offerings. Nanopore’s third-generation single-molecule sequencing represents a viable alternative to current commercial NGS-based PGT-A solutions for aneuploidy and segmental imbalance (≥10 Mb) screening of single-/multicell or trophectoderm biopsy samples.

6 March 2024 in Graefe's Archive for Clinical and Experimental Ophthalmology
Verified
Journal Article
Open access
PDF file
CWCarolyn Yu Tung Wong
Carolyn Yu Tung Wong
COCiara O’bYrne
Ciara O’bYrne
PTPriyal Taribagil
Priyal Taribagil
TLTiming Liu
Timing Liu
FAFares Antaki
Fares Antaki
PKPearse Keane
Pearse Keane
Aim: Code-free deep learning (CFDL) allows clinicians without coding expertise to build high-quality artificial intelligence (AI) models without writing code. In this review, we comprehensively review the advantages that CFDL offers over bespoke expert-designed deep learning (DL). As exemplars, we use the following tasks: (1) diabetic retinopathy screening, (2) retinal multi-disease classification, (3) surgical video classification, (4) oculomics and (5) resource management. Methods: We performed a search for studies reporting CFDL applications in ophthalmology in MEDLINE (through PubMed) from inception to June 25, 2023, using the keywords ‘autoML’ AND ‘ophthalmology’. After identifying 5 CFDL studies looking at our target tasks, we performed a subsequent search to find corresponding bespoke DL studies focused on the same tasks. Only English-written articles with full text available were included. Reviews, editorials, protocols and case reports or case series were excluded. We identified ten relevant studies for this review. Results: Overall, studies were optimistic towards CFDL’s advantages over bespoke DL in the five ophthalmological tasks. However, much of such discussions were identified to be mono-dimensional and had wide applicability gaps. High-quality assessment of better CFDL applicability over bespoke DL warrants a context-specific, weighted assessment of clinician intent, patient acceptance and cost-effectiveness. We conclude that CFDL and bespoke DL are unique in their own assets and are irreplaceable with each other. Their benefits are differentially valued on a case-to-case basis. Future studies are warranted to perform a multidimensional analysis of both techniques and to improve limitations of suboptimal dataset quality, poor applicability implications and non-regulated study designs. Conclusion: For clinicians without DL expertise and easy access to AI experts, CFDL allows the prototyping of novel clinical AI systems. CFDL models concert with bespoke models, depending on the task at hand. A multidimensional, weighted evaluation of the factors involved in the implementation of those models for a designated task is warranted.

25 August 2023 in medRxiv
Verified
Preprint
Open access
PDF file
TLTiming Liu
Timing Liu
ASAlagu Sankareswaran
Alagu Sankareswaran
GPGordon Paterson
Gordon Paterson
GTGenes & Health Research Team
Genes & Health Research Team
DFDiane P Fraser
Diane P Fraser
SFSarah Finer
Sarah Finer
Aims: Correct classification of type 1 (T1D) and type 2 diabetes (T2D) is challenging due to overlapping clinical features and the increasingly early onset of T2D, particularly in South Asians. We used polygenic risk scores (PRSs) in a British Bangladeshi and Pakistani population with diabetes to estimate the proportion and misclassification rate of T1D in insulin-treated individuals with ambiguous features. Methods: Using linked health records from the Genes & Health cohort (n=38,344) we defined four groups: 31 T1D cases, 1,842 T2D cases, and after excluding these, 839 insulin-treated individuals with ambiguous features and 5,174 controls. Combining these with 307 confirmed T1D cases and 307 controls from India, we calculated ancestry-corrected PRSs for T1D and T2D, with which we estimated the proportion of T1D cases within the ambiguous group and evaluated misclassification. Results: We estimated that the prevalence of T1D was ∼6% within the ambiguous group, or ∼4.5% within the subset who had T2D codes in their health records. We saw no significant association between the T1D or T2D PRS and BMI at diagnosis, time to insulin, or the presence of T1D or T2D diagnostic codes amongst the T2D or ambiguous cases, suggesting that these clinical features are not particularly helpful at aiding diagnosis in ambiguous cases. Conclusions: We estimate that about one in twenty of British Pakistanis and Bangladeshis with diabetes who are treated with insulin and have ambiguous clinical features have been classified incorrectly in their health records, and in fact have T1D. This emphasises that robust identification of T1D cases and appropriate clinical care may require routine measurement of diabetes autoantibodies and C-peptide. Research in context: What is already known about this subject?: - Correct classification of type 1 (T1D) and type 2 diabetes (T2D) is challenging due to overlapping clinical features and the increasingly early onset of T2D, particularly in people of South Asian descent. - Polygenic risk scores (PRSs) are useful tools to aid the classification of people with diabetes. What is the key question?: - What proportion of insulin-treated diabetic individuals with ambiguous clinical features have been clinically misclassified and in fact have T1D, amongst a cohort of British Pakistani and Bangladeshi adults? What are the new findings?: - Based on analyses of polygenic risk scores, the prevalence of T1D was found to be ∼6% within patients who were insulin-treated but with ambiguous features, and ∼4.5% were estimated to have been misclassified. - Clinical features such as BMI at diagnosis, time to insulin, or presence of T1D/T2D codes were not significantly associated with T1D or T2D PRS. How might this impact on clinical practice in the foreseeable future?: - These findings emphasise the importance of routine collection of diabetes autoantibodies and C-peptide measurements to identify T1D cases robustly, especially in countries where diabetes cases are diagnosed in primary care without input from diabetologists.

12 March 2024 in BJU International
Verified
Journal Article
Closed access
MAMahmoud Abuelnaga
Mahmoud Abuelnaga
RGRasha Gendy
Rasha Gendy
MYMark Yao
Mark Yao
CPCurtis Phelan
Curtis Phelan
JPJulie Pain
Julie Pain
BEBen Eddy
Ben Eddy