Ohio State Navbar

The Ohio State UniversityOffice of International Affairs

give a donation

Automatic segmentation of epidermis region from GVHD microscopy images

Research Scholar

Li Mao, BMI - School of Information and Control Engineering (China)
Khalid Niazi, Co-Researcher
Mikhail Viznyuk, Co-Researcher
Camille Elkins, Co-Researcher
Gerard Lozanski, Co-Researcher
Metin N. Gurcan, Faculty Mentor


Li Mao has been with the Clinical Image Analysis Lab as a visiting scholar since February 2012. His visit is funded by the Xi'an University of Architecture and Technology (XAUAT) and the China State Administration of Foreign Experts Affairs. He received his master's degree in computer science and technology from XAUAT in 2006, and is now a lecturer at that university's School of Information and Control Engineering. His research interests include digital watermarking and digital image processing. He has had seven publications featured in national and international journals and conferences, five as first author.

What is the issue or problem addressed in your research?

Graft-versus-host disease (GVHD) is a complication that can occur after a stem cell or bone marrow transplant if the newly transplanted donor cells attack the transplant recipient's body. The diagnosis often requires a pathologist to review pathological specimens under the microscope. Presence of necrotic keratinocytes in the epidermis region is often regarded as one of the symptoms of GVHD. We use computerized image analysis to accurately determine the epidermis boundaries from the scanned images of skin biopsy. Automated delineation of epidermis region is a challenging task due to several reasons: a) low contrast between the epidermis and dermis, b) irregular and fuzzy epidermis borders, c) artifacts such as skin texture, air bubbles and hair, and d) non-uniform coloring inside the epidermis and dermis.

What methodology did you use in your research?

Our proposed mathematical framework can be divided into three steps:

(i) Image preprocessing: Color normalization is necessary to reduce the variations in scanned images of skin biopsy.

(ii) Image segmentation: The adaptive clustering method attempts to decompose the skin image data into set of disjoint clusters (epidermis and dermis) by optimizing a cost function.

(iii) Segmentation evaluation: The computer output is compared with the manual ground truth using Dice's coefficient as a score.

What are the purpose/rationale and implications of your research?

The purpose of the research is to present an automated color segmentation method to delineate the epidermis region from the digital images of skin biopsy. The preliminary results of our efforts are quite promising. Currently, we are working to improve the accuracy and robustness of our method. Once fully developed, the method can also be useful for several other skin diseases and can help pathologists, dermatologists to accurately and rapidly analyze large quantities of skin biopsies.