student profile: Mr Dingqian Wang


Map

Thesis work

Thesis title: Multi-parametric imaging data processing analysis for brain tumour diagnose

Supervisors: Xiu WANG

Thesis abstract:

�p�Gliomas comprise about 30% of all primary central nervous system (CNS) tumours and 80% of all malignant brain tumours. The whole slide images have been regarded as the gold standard in the diagnosis of brain tumours hundreds of years ago. Accurate grading based on pathology images is crucial, as the therapeutic strategies are disparate for different grades, which may further influence the patient’s prognosis. However, molecular subsets of gliomas behave in biologically distinct ways. Their locations in the brain, rates of growth, and responses to therapy differ with their genotypes. After that, we have focused on brain tumour subtyping including LGGs can be grouped into three robust molecular classed on the basis of mutations in isocitrate dehydrogenase and co-deletion of 1p and 19q; GBMs can be grouped into two robust molecular classed on the basis of mutation in isocitrate dehydrogenase. Recently, much effort has been focused on the neuro-pathology diagnosis by Machine learning. The aim of the present study was to use machine learning for the glioma grading through evaluation of the cell morphological character and proliferation marker Ki-67. We developed a machine learning modular pipeline for the classification of 116 patients with glioma grade II, III, IV by digital pathology images. The primary results indicated that this modular pipeline approach has a high diagnostic accuracy from limited data availability and giving the flexibility to be used within brain tumour grading of machine learning applications. This machine learning based neural network classification algorithm achieved more than 93% accuracy on validation data set and produced 97%, 83%, 96% in GBM, Grade III, and grade II respectively. Furthermore, we used Local Interpretable Model-Agnostic Explanations (LIME) algorithm to reveal the features underlying grading prediction with a reasonable explanation for grading results. We think explaining an individual prediction output from the machine learning model is an effective way of assessing trust. As an efficient tool, LIME is much easier to approximate a black-box model by a simple model locally to facilitate such trust for machine learning practitioners and a good choice for pathology diagnosis. These results may provide a new view for clinicians for the glioma grading system and can work as a new platform to address more tumour-grading task by leveraging the combined machine learning modular.�/p�

Selected publications

Download citations: PDF RTF Endnote

Journals

  • Wang, X., Wang, D., Yao, Z., Xin, B., Wang, B., Lan, C., Qin, Y., Xu, S., He, D., Liu, Y. (2019). Machine Learning Models for Multiparametric Glioma Grading with Quantitative Result Interpretations. Frontiers in Neuroscience, 12(January 2019). [More Information]

2019

  • Wang, X., Wang, D., Yao, Z., Xin, B., Wang, B., Lan, C., Qin, Y., Xu, S., He, D., Liu, Y. (2019). Machine Learning Models for Multiparametric Glioma Grading with Quantitative Result Interpretations. Frontiers in Neuroscience, 12(January 2019). [More Information]

Note: This profile is for a student at the University of Sydney. Views presented here are not necessarily those of the University.