Medical Informatics

Medical informatics is the area of information processing, utilizing computational approaches to achieve better understanding of many different forms of medical information, in the hope that it could be useful in improving human health. I conducted following projects in the past, but open to any related research topics.

Mammogram analysis

Breast cancer is one of the most fatal type of cancer in women as reported by American Cancer Society. Like other types of cancers, early diagnosis of malicious tumors is the critical factor in overcoming the breast cancer. Breast cancer screening, has been used for an attempt to accomplish the early detection by periodic check up using mammogram (x-ray), ultrasound, or MRI. With the recent advances in digital mammography devices, several key issues have arisen for improving the early breast cancer detection using the digital mammograms:

First, with the massive amount of women in need of annual or bi-annual check up, how can computational image processing and machine learning techniques can help detect suspicious regions of massive amount of high resolution mammogram data, accurately?

Second, growing number of mammogram testing and false positive detections of suspicious regions increase the healthcare cost both for patients and the insurance plans. Can computational approaches reduce these false positive rates and reduce the unnecessary medical expenditures?

Third, there are different views (CC, MLO) and different devices (X-ray, sonar, MRI) that can achieve diagnostic imaging for breasts. Can we utilize multiple sources of information to improve performance or enhance the understanding of the images?

Microarray analysis

DNA Microarrays, is a way of genetic information analysis. Statistical methods have been applied in this domain to figure out which (set of) genes are important to discriminate the cancerous and non-cancerous genes. The problem is very challenging as there are so many genes (features) compared to relatively small number of patient data (samples). Thus, many feature selection methods in machine learning were developed and applied for this domain. The problem is still open and many other applications can be possible.