Chuan Lu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Chuan Lu has a total of 135 co-authors. Here are the top co-authors by number of shared publications:
- George Condous
Department of Obstetrics and Gynaecology, Nepean Hospital, Kingswood, NSW 2747, Australia
10 shared publications - Shannon Reid
Department of Obstetrics and Gynaecology, Liverpool Hospital, Liverpool, NSW, Australia
8 shared publications B. Shakeri
Acute Gynecology, Early Pregnancy & Advanced Endoscopic Surgery Unit, Sydney Medical School Nepean University of Sydney Nepean Hospital Penrith, Sydney Australia
5 shared publications- Mercedes Espada
Department of Obstetrics and Gynaecology, Nepean Hospital, Kingswood, NSW 2747, Australia
5 shared publications - Batool Nadim
Hawkesbury Hospital, Norwest Private Hospital The University of Notre Dame Australia, University of Sydney Sydney New South Wales Australia
3 shared publications I. Casikar
Acute Gynecology, Early Pregnancy & Advanced Endoscopic Surgery Unit, Sydney Medical School Nepean University of Sydney Nepean Hospital Penrith, Sydney Australia
3 shared publications
Latest Publications
- Open Access
- 16 October 2026
- Published by Elsevier in SSRN Electronic Journal
GIST: A Spatio-Temporal Graph Framework for Interpretable EEG Analysis
Background and objective: Electroencephalogram (EEG) analysis requires methods that can jointly model functional connectivity and temporal dynamics while still supporting inspection of decision-relevant patterns. This study proposes the GAT-Inspired Spatio-Temporal (GIST) framework as a graph-based method for EEG classification with built-in attention-based interpretability.Methods: GIST represents each EEG segment as a sequence of window-level graphs constructed from domain-derived node descriptors and coherence-based inter-channel connectivity. The framework combines edge attention, node attention, and temporal attention within a unified graph-based architecture, enabling prediction-linked inspection of pairwise connectivity, electrode relevance, and temporal salience. The same architecture was evaluated across three public EEG settings covering stimulation site identification, neonatal pain detection, and emotion recognition.Results: GIST achieved 100% accuracy (AUC 1.00) for stimulation site identification in the controlled TES setting, 75.5% accuracy (AUC 0.771) for neonatal pain detection, and 82.6% accuracy (AUC 0.918) for emotion recognition. In addition to predictive performance, the framework supported quantitative aggregation and visualisation of edge-, node-, and temporal-attention scores, enabling qualitative interpretation of task-relevant patterns.Conclusions: Across complementary EEG tasks with different recording conditions and objectives, GIST provided a consistent graph-based analysis framework that combined competitive classification performance with interpretable attention-derived relevance patterns. These results support its use as a reusable methodological scaffold for spatio-temporal EEG analysis.
- 16 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessGIST: A Spatio-Temporal Graph Framework for Interpretable EEG Analysis
Background and objective: Electroencephalogram (EEG) analysis requires methods that can jointly model functional connectivity and temporal dynamics while still supporting inspection of decision-relevant patterns. This study proposes the GAT-Inspired Spatio-Temporal (GIST) framework as a graph-based method for EEG classification with built-in attention-based interpretability.Methods: GIST represents each EEG segment as a sequence of window-level graphs constructed from domain-derived node descriptors and coherence-based inter-channel connectivity. The framework combines edge attention, node attention, and temporal attention within a unified graph-based architecture, enabling prediction-linked inspection of pairwise connectivity, electrode relevance, and temporal salience. The same architecture was evaluated across three public EEG settings covering stimulation site identification, neonatal pain detection, and emotion recognition.Results: GIST achieved 100% accuracy (AUC 1.00) for stimulation site identification in the controlled TES setting, 75.5% accuracy (AUC 0.771) for neonatal pain detection, and 82.6% accuracy (AUC 0.918) for emotion recognition. In addition to predictive performance, the framework supported quantitative aggregation and visualisation of edge-, node-, and temporal-attention scores, enabling qualitative interpretation of task-relevant patterns.Conclusions: Across complementary EEG tasks with different recording conditions and objectives, GIST provided a consistent graph-based analysis framework that combined competitive classification performance with interpretable attention-derived relevance patterns. These results support its use as a reusable methodological scaffold for spatio-temporal EEG analysis.
- Open Access
- 11 June 2025
- Published by Elsevier in Computers and Electronics in Agriculture
- Vol. 237
DeepCanola: Phenotyping brassica pods using semi-synthetic data and active learning
- 11 June 2025
- Published by Elsevier in Computers and Electronics in Agriculture
- Vol. 237
Open AccessDeepCanola: Phenotyping brassica pods using semi-synthetic data and active learning
- Open Access
- 6 January 2025
- Published by Oxford University Press (OUP) in GigaScience
- Vol. 14
Unlocking the power of AI for phenotyping fruit morphology in Arabidopsis
Deep learning can revolutionise high-throughput image-based phenotyping by automating the measurement of complex traits, a task that is often labour-intensive, time-consuming, and prone to human error. However, its precision and adaptability in accurately phenotyping organ-level traits, such as fruit morphology, remain to be fully evaluated. Establishing the links between phenotypic and genotypic variation is essential for uncovering the genetic basis of traits and can also provide an orthologous test of pipeline effectiveness. In this study, we assess the efficacy of deep learning for measuring variation in fruit morphology in Arabidopsis using images from a multiparent advanced generation intercross (MAGIC) mapping family. We trained an instance segmentation model and developed a pipeline to phenotype Arabidopsis fruit morphology, based on the model outputs. Our model achieved strong performance with an average precision of 88.0% for detection and 55.9% for segmentation. Quantitative trait locus analysis of the derived phenotypic metrics of the MAGIC population identified significant loci associated with fruit morphology. This analysis, based on automated phenotyping of 332,194 individual fruits, underscores the capability of deep learning as a robust tool for phenotyping large populations. Our pipeline for quantifying pod morphological traits is scalable and provides high-quality phenotype data, facilitating genetic analysis and gene discovery, as well as advancing crop breeding research.
- 6 January 2025
- Published by Oxford University Press (OUP) in GigaScience
- Vol. 14
Open AccessUnlocking the power of AI for phenotyping fruit morphology in Arabidopsis
Deep learning can revolutionise high-throughput image-based phenotyping by automating the measurement of complex traits, a task that is often labour-intensive, time-consuming, and prone to human error. However, its precision and adaptability in accurately phenotyping organ-level traits, such as fruit morphology, remain to be fully evaluated. Establishing the links between phenotypic and genotypic variation is essential for uncovering the genetic basis of traits and can also provide an orthologous test of pipeline effectiveness. In this study, we assess the efficacy of deep learning for measuring variation in fruit morphology in Arabidopsis using images from a multiparent advanced generation intercross (MAGIC) mapping family. We trained an instance segmentation model and developed a pipeline to phenotype Arabidopsis fruit morphology, based on the model outputs. Our model achieved strong performance with an average precision of 88.0% for detection and 55.9% for segmentation. Quantitative trait locus analysis of the derived phenotypic metrics of the MAGIC population identified significant loci associated with fruit morphology. This analysis, based on automated phenotyping of 332,194 individual fruits, underscores the capability of deep learning as a robust tool for phenotyping large populations. Our pipeline for quantifying pod morphological traits is scalable and provides high-quality phenotype data, facilitating genetic analysis and gene discovery, as well as advancing crop breeding research.
- 1 February 2024
- Published by Springer Nature
- p. 286-298
Graph Attention Based Spatial Temporal Network for EEG Signal Representation
- 1 February 2024
- Published by Springer Nature
- p. 286-298
Graph Attention Based Spatial Temporal Network for EEG Signal Representation
- Top Cited PapersOpen Access
- 28 October 2021
- Published by Elsevier in The Innovation
- Vol. 2 (4) , 100179
Artificial intelligence: A powerful paradigm for scientific research
- 28 October 2021
- Published by Elsevier in The Innovation
- Vol. 2 (4) , 100179
Top Cited PapersOpen AccessArtificial intelligence: A powerful paradigm for scientific research
- Open Access
- 1 March 2020
- Published by Oxford University Press (OUP) in GigaScience
- Vol. 9 (3)
DeepPod: a convolutional neural network based quantification of fruit number in Arabidopsis
High-throughput phenotyping based on non-destructive imaging has great potential in plant biology and breeding programs. However, efficient feature extraction and quantification from image data remains a bottleneck that needs to be addressed. Advances in sensor technology have led to the increasing use of imaging to monitor and measure a range of plants including the model Arabidopsis thaliana. These extensive datasets contain diverse trait information, but feature extraction is often still implemented using approaches requiring substantial manual input. The computational detection and segmentation of individual fruits from images is a challenging task, for which we have developed DeepPod, a patch-based 2-phase deep learning framework. The associated manual annotation task is simple and cost-effective without the need for detailed segmentation or bounding boxes. Convolutional neural networks (CNNs) are used for classifying different parts of the plant inflorescence, including the tip, base, and body of the siliques and the stem inflorescence. In a post-processing step, different parts of the same silique are joined together for silique detection and localization, whilst taking into account possible overlapping among the siliques. The proposed framework is further validated on a separate test dataset of 2,408 images. Comparisons of the CNN-based prediction with manual counting (R2 = 0.90) showed the desired capability of methods for estimating silique number. The DeepPod framework provides a rapid and accurate estimate of fruit number in a model system widely used by biologists to investigate many fundemental processes underlying growth and reproduction
- 1 March 2020
- Published by Oxford University Press (OUP) in GigaScience
- Vol. 9 (3)
Open AccessDeepPod: a convolutional neural network based quantification of fruit number in Arabidopsis
High-throughput phenotyping based on non-destructive imaging has great potential in plant biology and breeding programs. However, efficient feature extraction and quantification from image data remains a bottleneck that needs to be addressed. Advances in sensor technology have led to the increasing use of imaging to monitor and measure a range of plants including the model Arabidopsis thaliana. These extensive datasets contain diverse trait information, but feature extraction is often still implemented using approaches requiring substantial manual input. The computational detection and segmentation of individual fruits from images is a challenging task, for which we have developed DeepPod, a patch-based 2-phase deep learning framework. The associated manual annotation task is simple and cost-effective without the need for detailed segmentation or bounding boxes. Convolutional neural networks (CNNs) are used for classifying different parts of the plant inflorescence, including the tip, base, and body of the siliques and the stem inflorescence. In a post-processing step, different parts of the same silique are joined together for silique detection and localization, whilst taking into account possible overlapping among the siliques. The proposed framework is further validated on a separate test dataset of 2,408 images. Comparisons of the CNN-based prediction with manual counting (R2 = 0.90) showed the desired capability of methods for estimating silique number. The DeepPod framework provides a rapid and accurate estimate of fruit number in a model system widely used by biologists to investigate many fundemental processes underlying growth and reproduction
- 9 January 2020
- Published by Wiley in Australian and New Zealand Journal of Obstetrics and Gynaecology
- Vol. 60 (2) , 258-263
Ultrasound evaluation of pouch of Douglas obliteration and rectal deep endometriosis in women who have had previous combined colorectal and gynaecological laparoscopic surgery for rectal endometriosis: A pilot study
Background: Ultrasound has been demonstrated to accurately diagnose rectal deep endometriosis (DE) and pouch of Douglas (POD) obliteration. The role of ultrasound in the assessment of patients who have undergone surgery for rectal DE and POD obliteration has not been evaluated.Aim: To describe the transvaginal ultrasound (TVS) findings of patients who have undergone rectal surgery for DE.Materials and Methods: An observational cross‐sectional study at a tertiary care centre in Sydney, Australia between January and April 2017. Patients previously treated for rectal DE (low anterior resection vs rectal shaving/disc excision) were recruited and asked to complete a questionnaire on their current symptoms. On TVS, POD state and rectal DE were assessed. Correlating recurrence of POD obliteration and/or rectal DE to surgery type and symptoms was done.Results: Fifty‐six patients were contacted; 22/56 (39.3%) attended for the study visit. Average interval of surgery to study visit was 52.8 ± 24.6 months. Surgery type breakdown was as follows: low anterior resection (56%) and rectal shaving/disc excision (44%). The prevalence of POD obliteration was 16/22 (72.7%) intraoperatively and 8/22 (36.4%) at study visit, as per the sliding sign. Nine patients (39.1%) had evidence on TVS of recurrent rectal DE. Recurrence of POD obliteration and rectal DE was not associated with surgery type or symptomatology.Conclusion: Despite surgery for rectal DE, many patients have a negative sliding sign on TVS, representing POD obliteration, and rectal DE. Our numbers are too small to correlate with the surgery type or their current symptoms.
- 9 January 2020
- Published by Wiley in Australian and New Zealand Journal of Obstetrics and Gynaecology
- Vol. 60 (2) , 258-263
Ultrasound evaluation of pouch of Douglas obliteration and rectal deep endometriosis in women who have had previous combined colorectal and gynaecological laparoscopic surgery for rectal endometriosis: A pilot study
Background: Ultrasound has been demonstrated to accurately diagnose rectal deep endometriosis (DE) and pouch of Douglas (POD) obliteration. The role of ultrasound in the assessment of patients who have undergone surgery for rectal DE and POD obliteration has not been evaluated.Aim: To describe the transvaginal ultrasound (TVS) findings of patients who have undergone rectal surgery for DE.Materials and Methods: An observational cross‐sectional study at a tertiary care centre in Sydney, Australia between January and April 2017. Patients previously treated for rectal DE (low anterior resection vs rectal shaving/disc excision) were recruited and asked to complete a questionnaire on their current symptoms. On TVS, POD state and rectal DE were assessed. Correlating recurrence of POD obliteration and/or rectal DE to surgery type and symptoms was done.Results: Fifty‐six patients were contacted; 22/56 (39.3%) attended for the study visit. Average interval of surgery to study visit was 52.8 ± 24.6 months. Surgery type breakdown was as follows: low anterior resection (56%) and rectal shaving/disc excision (44%). The prevalence of POD obliteration was 16/22 (72.7%) intraoperatively and 8/22 (36.4%) at study visit, as per the sliding sign. Nine patients (39.1%) had evidence on TVS of recurrent rectal DE. Recurrence of POD obliteration and rectal DE was not associated with surgery type or symptomatology.Conclusion: Despite surgery for rectal DE, many patients have a negative sliding sign on TVS, representing POD obliteration, and rectal DE. Our numbers are too small to correlate with the surgery type or their current symptoms.
- 1 September 2019
- Published by Wiley in Journal of Ultrasound in Medicine
- Vol. 38 (9) , 2437-2445
Doppler Color Scoring System in Women With an Incomplete Miscarriage: Interobserver and Intraobserver Reproducibility Study
Doppler Color Scoring (DCS) has been used to predict successful expectant management of incomplete miscarriage. The aim of this study was to assess inter- and intraobserver reproducibility of the DCS system in women with incomplete miscarriage noted on transvaginal sonography. This was a prospective reproducibility study involving offline analysis of 32 prerecorded video sets on transvaginal sonography in real time of women with incomplete miscarriage. Vascularization of retained products of conception was recorded using the DCS system adopted from the International Ovarian Tumor Analysis group. Five gynecologic sonologists of varying experience assigned a DCS classification to each video in the analysis. The same videos were reanalyzed, in a different order, at least 7 days later, to assess intraobserver agreement. Inter- and intraobserver correlations were performed to determine agreement. Interobserver agreement was also measured between each observer and the reference standard (G.C.). A Cohen's κ coefficient value less than 0 suggests poor agreement, 0.01 to 0.20 slight, 0.21 to 0.40 fair, 0.41 to 0.60 moderate, 0.61 to 0.80 substantial, and 0.81 and 0.99 almost perfect. Interobserver agreement for all observers for DCS allocation ranged from 0.480 to 0.751. Overall interobserver agreement for 5 observers was substantial (κ, 0.626). Overall interobserver agreements for the 2 inexperienced and 3 experienced observers compared to G.C. were 0.521 and 0.618, respectively. Experienced observers achieved overall almost perfect intraobserver agreement, compared to substantial agreement for inexperienced sonologists. DCS interobserver reproducibility between all observers and GC ranged from moderate to substantial. DCS intraobserver reproducibility was substantial to almost perfect. The DCS system appears to be a reproducible tool in evaluating women with incomplete miscarriage.
- 1 September 2019
- Published by Wiley in Journal of Ultrasound in Medicine
- Vol. 38 (9) , 2437-2445
Doppler Color Scoring System in Women With an Incomplete Miscarriage: Interobserver and Intraobserver Reproducibility Study
Doppler Color Scoring (DCS) has been used to predict successful expectant management of incomplete miscarriage. The aim of this study was to assess inter- and intraobserver reproducibility of the DCS system in women with incomplete miscarriage noted on transvaginal sonography. This was a prospective reproducibility study involving offline analysis of 32 prerecorded video sets on transvaginal sonography in real time of women with incomplete miscarriage. Vascularization of retained products of conception was recorded using the DCS system adopted from the International Ovarian Tumor Analysis group. Five gynecologic sonologists of varying experience assigned a DCS classification to each video in the analysis. The same videos were reanalyzed, in a different order, at least 7 days later, to assess intraobserver agreement. Inter- and intraobserver correlations were performed to determine agreement. Interobserver agreement was also measured between each observer and the reference standard (G.C.). A Cohen's κ coefficient value less than 0 suggests poor agreement, 0.01 to 0.20 slight, 0.21 to 0.40 fair, 0.41 to 0.60 moderate, 0.61 to 0.80 substantial, and 0.81 and 0.99 almost perfect. Interobserver agreement for all observers for DCS allocation ranged from 0.480 to 0.751. Overall interobserver agreement for 5 observers was substantial (κ, 0.626). Overall interobserver agreements for the 2 inexperienced and 3 experienced observers compared to G.C. were 0.521 and 0.618, respectively. Experienced observers achieved overall almost perfect intraobserver agreement, compared to substantial agreement for inexperienced sonologists. DCS interobserver reproducibility between all observers and GC ranged from moderate to substantial. DCS intraobserver reproducibility was substantial to almost perfect. The DCS system appears to be a reproducible tool in evaluating women with incomplete miscarriage.
- 15 May 2019
- Published by Wiley in Journal of Ultrasound in Medicine
- Vol. 38 (12) , 3301-3309
A Novel Ultrasound Technique Called Saline Infusion SonoPODography to Visualize and Understand the Pouch of Douglas and Posterior Compartment Contents: A Feasibility Study
Women with infertility are often investigated with saline‐infusion sonohysterography and hysterosalpingo‐contrast‐sonography. The high prevalence of endometriosis in this population also warrants an evaluation with transvaginal ultrasound for deep endometriosis. To minimize investigations, we prospectively evaluated the feasibility of a novel combined ultrasound technique called saline‐infusion sonoPODography. In most patients, the fluid infused to assess the cavity and tubal patency spilled through patent tubes and filled the pouch of Douglas, yielding a “standoff” view of posterior compartment structures, including uterosacral ligaments, rectovaginal septum, and the pouch of Douglas. We believe this improved our ability to evaluate this space.
- 15 May 2019
- Published by Wiley in Journal of Ultrasound in Medicine
- Vol. 38 (12) , 3301-3309
A Novel Ultrasound Technique Called Saline Infusion SonoPODography to Visualize and Understand the Pouch of Douglas and Posterior Compartment Contents: A Feasibility Study
Women with infertility are often investigated with saline‐infusion sonohysterography and hysterosalpingo‐contrast‐sonography. The high prevalence of endometriosis in this population also warrants an evaluation with transvaginal ultrasound for deep endometriosis. To minimize investigations, we prospectively evaluated the feasibility of a novel combined ultrasound technique called saline‐infusion sonoPODography. In most patients, the fluid infused to assess the cavity and tubal patency spilled through patent tubes and filled the pouch of Douglas, yielding a “standoff” view of posterior compartment structures, including uterosacral ligaments, rectovaginal septum, and the pouch of Douglas. We believe this improved our ability to evaluate this space.
- 14 July 2018
- Published by Wiley in Acta Obstetricia et Gynecologica Scandinavica
- Vol. 97 (11) , 1287-1292
To determine the optimal ultrasonographic screening method for rectal/rectosigmoid deep endometriosis: Ultrasound “sliding sign,” transvaginal ultrasound direct visualization or both?
Introduction: The study aim was to evaluate the transvaginal sonography (TVS) “sliding sign” alone, direct visualization of the bowel with TVS, and the combination of both methods (ie “sliding sign” and direct visualization of the bowel), to determine the optimal TVS method for the prediction of rectal/rectosigmoid deep endometriosis (DE).Material and methods: Multicenter prospective observational study (January 2009‐February 2017). All women underwent TVS to determine whether the “sliding sign” was positive/negative and whether rectal/rectosigmoid DE was present, followed by laparoscopic surgery. The association between a negative TVS “sliding sign” alone and the direct visualization of a rectal/rectosigmoid DE nodule alone during the TVS were correlated with the presence of rectal/rectosigmoid DE at laparoscopy. Accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and likelihood ratios (LRs) were evaluated. Data were analyzed using Fisher's exact test.Results: During the recruitment period, 410 consecutive women with suspected endometriosis were included. Complete TVS and laparoscopic surgical outcomes were available for 376 of the women (91.7%). Complete TVS and laparoscopic data were available for 376 women. Of the 376 women 76 (20.2%) had rectal/rectosigmoid DE at laparoscopy. The accuracy, sensitivity, specificity, PPV, NPV, positive and negative LRs for each method to predict bowel DE were: negative “sliding sign”: 87%, 73.7%, 90.3%, 65.9%, 93.1%, 7.62, and 0.29, respectively; direct visualization: 91.0%, 86.8%, 92.3%, 74.2%, 96.5%, 11.3, and 0.14, respectively; combined approach: 90.2%, 69.7%, 95.3%, 79.1%, 92.6%, 14.94, and 0.32, respectively. A negative TVS “sliding sign” was significantly associated with the need for bowel surgery (P < 0.05).Conclusions: The combination of the TVS “sliding sign” and direct visualization of the bowel during TVS appears to provide the most accurate assessment for the identification of rectal/rectosigmoid DE preoperatively.
- 14 July 2018
- Published by Wiley in Acta Obstetricia et Gynecologica Scandinavica
- Vol. 97 (11) , 1287-1292
To determine the optimal ultrasonographic screening method for rectal/rectosigmoid deep endometriosis: Ultrasound “sliding sign,” transvaginal ultrasound direct visualization or both?
Introduction: The study aim was to evaluate the transvaginal sonography (TVS) “sliding sign” alone, direct visualization of the bowel with TVS, and the combination of both methods (ie “sliding sign” and direct visualization of the bowel), to determine the optimal TVS method for the prediction of rectal/rectosigmoid deep endometriosis (DE).Material and methods: Multicenter prospective observational study (January 2009‐February 2017). All women underwent TVS to determine whether the “sliding sign” was positive/negative and whether rectal/rectosigmoid DE was present, followed by laparoscopic surgery. The association between a negative TVS “sliding sign” alone and the direct visualization of a rectal/rectosigmoid DE nodule alone during the TVS were correlated with the presence of rectal/rectosigmoid DE at laparoscopy. Accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and likelihood ratios (LRs) were evaluated. Data were analyzed using Fisher's exact test.Results: During the recruitment period, 410 consecutive women with suspected endometriosis were included. Complete TVS and laparoscopic surgical outcomes were available for 376 of the women (91.7%). Complete TVS and laparoscopic data were available for 376 women. Of the 376 women 76 (20.2%) had rectal/rectosigmoid DE at laparoscopy. The accuracy, sensitivity, specificity, PPV, NPV, positive and negative LRs for each method to predict bowel DE were: negative “sliding sign”: 87%, 73.7%, 90.3%, 65.9%, 93.1%, 7.62, and 0.29, respectively; direct visualization: 91.0%, 86.8%, 92.3%, 74.2%, 96.5%, 11.3, and 0.14, respectively; combined approach: 90.2%, 69.7%, 95.3%, 79.1%, 92.6%, 14.94, and 0.32, respectively. A negative TVS “sliding sign” was significantly associated with the need for bowel surgery (P < 0.05).Conclusions: The combination of the TVS “sliding sign” and direct visualization of the bowel during TVS appears to provide the most accurate assessment for the identification of rectal/rectosigmoid DE preoperatively.