Approved Research

Artificial intelligence (AI) and radiomics for cardiac and brain magnetic resonance based imaging analysis: artifact reduction, quality assessment and interaction between cardiac and brain

Lay summary

Cardiac and Brain magnetic resonance is a high-quality and non-invasive imaging tool to study heart and brain structure along with function.

Benefit from fast-developing computer technology, artificial intelligence (AI) has become more and more common in daily life. Deep learning (DL) as an important part of AI, showed excellent performance in many tasks. In the field of medical image analysis, DL already exhibited perfect results in some areas, however, its disadvantages were also obvious. One of the most import drawback is that DL is hard to explain. Meanwhile, another method, called 'radiomics', provide an more explicit way to analyse and explain medical images.

Therefore, in this study, I intend to use the combination of AI, radiomics to analyse CMR images. I will focus on the following questions: 1) detect low quality images and try to improve them; 2) design several automatic and accurate segmentation models to facilitate image analysis; 3) comparing models' performance and human-level performance; 4) extract and select radiomics features from images; 5) combining clinical information, radiomic features and DL models to study the interaction between brain and heart under different diseases; 5) finally, incorporate all parts together.

According to my observation and previous experiment on a small dataset. We hypothesized that combination of AI and radiomics could provides a more accurate and robust analysis framework .

This study will be conducted for 2.5-3 years with an excellent server. I hope that our proposed system could facilitate medical images quality control/improvement and clinical decision.

The codes for image processing, model architecture will be open-sourced with my paper and thesis. Also, I hope the manual segmenation of CMR images (me, 2 years of CMR experience) could be used in other research projects, and I will upload this part of data to UK Biobank if necessary.