Dr. Maxine Tan

Senior Lecturer
School of Engineering

maxine.tan@monash.edu
+603 5515 9702
Room 2-4-40
ORCID

Personal statement

Dr Maxine is an active researcher and have made important contributions in Computer-Aided Diagnosis (CAD) for breast cancer risk prediction, ovarian cancer prognosis (i.e., new methods to evaluate drug treatment response of ovarian cancer patients), and breast, lung and ovarian cancer detection and classification. She is also currently actively doing research in the field of deep learning applied to lung cancer diagnosis in computed tomography (CT) scans and for histopathological images.

Dr Maxine recently developed a new scheme to detect bilateral mammographic density asymmetry between left and right breasts, in order to generate a new short-term risk prediction score that yields significantly higher prediction accuracy at the individual level compared to current fixed prediction models that are based on lifetime risk. The new update to the breast cancer screening guideline issued by the American Cancer Society in October 2015 states that women 55 years and older should transition to biennial screening or have the opportunity to continue screening annually based on a qualified recommendation. Based on this new guideline, the new tool she has developed would help to better stratify the women that require annual screening from biennial screening. Furthermore, the estimated cost of mammography screening in the US is very costly (e.g., $7.8 billion in 2010). Thus, her research work in this area has potential both to save lives and reduce screening costs, and is of great importance.

Recently, Dr Maxine's research work in deep learning helped her team and herself to secure a fundamental research grant scheme (FRGS) grant related to lung cancer detection and classification in CT scans. She has also secured several local and international grants related to deep learning, radiomics and cancer risk prediction in medical images. Her expertise and contributions in cancer imaging research has been well recognized by peers in the field. Dr Maxine's research results have been published in peer-reviewed scientific journals in Medical Imaging such as IEEE Transactions on Medical Imaging, Annals of Biomedical Engineering, Medical Physics, Physics in Medicine and Biology, Artificial Intelligence in Medicine, and IEEE Transactions on Biomedical Engineering. She has published 44 peer-reviewed articles and has a Google h-index of 14. She was also awarded the Cum Laude (Best Poster) Award and the Honorable Mention Award for the Computer-Aided Diagnosis and Digital Pathology tracks, respectively of the SPIE Medical Imaging 2015 conference.

Academic degrees

  • Doctor of Philosophy in Engineering Sciences, Vrije Universiteit Brussel (VUB), 2012
  • Master in Electrical & Electronics Engineering (MEng), University of Nottingham, Malaysia Campus, 2006

Professional affiliations

Member of National Professional Bodies

  • Board of Engineers Malaysia (BEM) - Member

Research Interests

My research field is in medical imaging. For the past 11 years, I have been working to investigate and develop new quantitative imaging (QI) feature based clinical markers that can more effectively phenotype disease (e.g., cancer) risk, development and prognosis, which can thus assist in improving efficacy of disease screening, diagnosis and treatment. In current clinical practice, there is a steadily increasing trend of using more and more medical imaging examinations with high dimension and resolution data. For example, high resolution computed tomography (HRCT) with slice thickness smaller than 1mm has been routinely used to replace conventional X-ray radiography. These HRCT images carry considerable useful information that can phenotype biological and physical properties of the diseases. However, subjective image reading and interpretation by the radiologists are not able to accurately extract such important information due to the large inter-reader variability. As a result, deep learning in the medical image processing field as well as a new field of Radiomics has recently been attracting extensive research interest and effort. My research work in the past year as a Lecturer at Monash University, Malaysia Campus and in the previous three-and-a-half years as a Postdoc at the University of Oklahoma and University of Pittsburgh Medical Center (UPMC), Pittsburgh fits well with the goal of deep learning and Radiomics. I have participated in and carried out a number of pioneer works, which have never been investigated and reported by other research groups, in these new fields. The following are brief summaries of my recent contribution to this new field.

Research Projects

Title: Developed new deep learning based methods for lung cancer detection and diagnosis in computed tomography (CT) scans

Deep learning is becoming ubiquitous in the image recognition domain and has recently been shown to outperform conventional methods in various medical image processing and classification tasks. I have been working on improving benign/malignant classification and detection of lung cancers using deep neural networks. The deep neural networks will be incorporated in Computer-Aided Detection (CADe) and Computer-Aided Diagnosis (CADx) schemes to improve the accuracy and consistency among radiologists. I applied for the NVIDIA GPU Grant, which was approved by NVIDIA

Title: Developed and demonstrated the first near-term breast cancer risk assessment model and interactive computer-aided image display scheme

Developing a new personalized breast cancer screening paradigm for the Precision Medicine initiative requires new risk factors and prediction models to assess cancer risk of the individual women having imaging detectable breast cancer in the near term (e.g., the next annual screening) after a negative screening examination of interest. The current epidemiology based breast cancer risk prediction models, such as Gail and Tyrer-Cuzick typically estimate a long-term (or lifetime) risk of developing breast cancer in women. These risk prediction models do not have clinically-acceptable discriminatory power when applied to predict near-term cancer risk of the individual woman. In order to solve this clinically-important issue, I have been working to investigate new quantitative image feature based clinical markers, and also developed a new breast cancer risk prediction model and interface, whereby the user can easily upload a set of testing cases (images), and let the model compute a risk score (inside a yellow box on the upper-right corner of the display screen). The user can also view many middle/ intermediate computational results (maps) and the computed individual features that are used to build the final risk prediction model.

Title: Developed and reported the first global image feature based new CAD scheme of mammograms

Due to the higher false-positive detection rates, using current commercialized and region-based computer-aided detection (CAD) schemes of mammograms as a “second reader” failed to assist radiologists in improving accuracy in reading and interpreting mammograms. To resolve this issue, I developed and tested the first case-based CAD approach based on global bilateral image feature asymmetry on four-view (CC and MLO) mammograms. In order to select effective global image features used in the CAD scheme, I had also examined and modified a new and fast feature selection method, namely modified sequential floating forward selection (SFFS), which outperformed (i.e., achieved higher AUC results) and is more computationally efficient than other state-of-the-art feature selection methods. The new CAD scheme yielded a case-based area under the receiver operating characteristic curve (AUC) classification performance of 0.779±0.025 and a total classification accuracy rate of 72.3% on a large and diverse image dataset with 1896 FFDM screening examinations (812 positive for cancer and 1084 negative or benign). Our work demonstrated a new approach to develop CAD schemes of mammograms in the future. Our work and results are reported in several journal and conference papers.

Title: Developed and tested the first CAD scheme to automatically detect and track clinically relevant tumors from the CT images acquired pre- and post-chemotherapy

Currently, Response Evaluation Criteria in Solid Tumors (RECIST) guidelines are used to assess tumor response to chemotherapy, which has a number of limitations that lead to the low prediction accuracy on patients’ disease-free or overall survival rate. To solve this issue, I developed a unique deformable image registration based CAD scheme, and demonstrated for the first time that using this scheme enabled to automatically detect and track more tumors that are ignored and/or overlooked using current RECIST guidelines. Our preliminary studies indicated that using this new CAD scheme yielded significantly higher accuracy in predicting 6-month disease-free survival (DFS) of ovarian cancer patients who participated in the clinical trials to test new chemotherapy drugs. The study result was published in the IEEE Transactions on Medical Imaging journal

Title: New cancer prognosis assessment models for breast MRI and lung CT

I have also developed new CAD-based prediction models to better predict cancer prognosis based on the quantitative image features computed from medical images. For example, I helped to develop and test new models to predict breast tumor response to neoadjuvant chemotherapy using breast MR images and cancer recurrence risk of early stage lung cancer patients after tumor surgery treatment using CT images. In particular, my colleagues/collaborators and I compared quantitative CT lung tumor image features with two popular genomic biomarkers and demonstrated that using image features could not only yield higher prediction performance, but also provide supplementary information to the genomic biomarkers, so that fusion of image features with genomic biomarkers has great potential to further improve prediction performance.

Title: Developed new CAD schemes for breast mammography, lung CT and CT for assessing postmastectomy lymphedema

I developed new CAD schemes for breast cancer detection and diagnosis in mammograms as well as lung cancer detection in CT scans. I also developed a new histogram analysis based method to evaluate the CT scans of patients with post-mastectomy lymphedema.

Title: Deep learning applied to reinforcement learning

We are currently exploring different deep learning methods for unsupervised learning and reinforcement learning. These include training new deep learning models to control a robotic arm to pick-and-place objects.

Education

Units taught

ECE2111 - Signals and Systems

No content

Local grants

  • Investigation of New Imaging Markers to Automatically Detect and Classify Lung Cancer in Computed Tomography Scans, Maxine Tan (Principal Investigator), Boon Leong Lan, Kwan Hoong Ng, Wai Yee Chan, 2018-2021, Fundamental research grant scheme (FRGS)
  • Deep learning for palm seed classification, Maxine Tan (Co-Investigator), Mohammad Hisham Jaward, Vineetha Kallavally, 2018-2019, Monash University Malaysia Engineering-IT Collaborative Grant Scheme 2018, RM20,000
  • New quantitative imaging markers to predict breast cancer risk, Maxine Tan (Principal Investigator), 2017-2018, Monash University Malaysia Seed Grant, RM30,000
  • Prediction of coefficients in compressed content and its applications, Maxine Tan (Co-Investigator), Kok Sheik Wong, 2017-2018, Monash University Malaysia Collaborative Research Seed Grant, RM20,000

International grants

  • New imaging markers to automatically detect and classify lung cancer in computed tomography scans, Maxine Tan (Principal Investigator), 2017, NVIDIA GPU Grant, RM8,000

Current supervision

Mundher Al-Shabi

Lung Cancer Detection in CT Images Using Deep Learning

2017-2020

Monash University Malaysia

International Award/Recognition/Exhibition/Stewardship

  • Cum Laude (Best Poster) Award: A new CAD approach for improving efficacy of cancer screening - SPIE, 2015
  • Honorable Mention Award: An automated approach to improve the efficacy in detecting residual cancer cell for facilitating prognostic assessment of leukemia: an initial study - SPIE, 2015
  • News Story: Neural network cuts false-positive recalls - Medical Physics Web, 2014
  • News Story: New SVM model predicts near-term breast cancer risk - Health Imaging, 2013