Home :: Seed4

Statistical Modeling and Classification for Skin Cancer Detection using Hyperspectral and Microscopic Images

Vidya Manian Ph.D.


Skin cancer is the most common form of cancer that affects the outer layer of the skin. In this project we propose to (1) develop statistical models to discriminate between normal skin tissue and abnormal skin tissue (cancerous) from optical reflectance and transmittance properties of tissue using noninvasive hyperspectral imaging methods, (2) our models will be used with real data obtained from microscopy images in collaboration with the Center for Subsurface Sensing and Imaging Systems, Northeastern University (NU).

CenSSIS, NU will provide skin sample images which will be processed by the algorithms that will be implemented in this project. The algorithms include methods for shape, geometry and texture feature extraction and for classification. Feature extraction will be done using wavelet methods and Gabor filters to characterize skin structural properties and discriminate normal skin samples from abnormal (carcinoma or melanoma) skin samples. The experimental setup for the statistical model will make use of the CCD camera imaging system in the Laboratory for Applied Remote Sensing and Image Processing (LARSIP) for capturing hyperspectral images.

The images modeled will have clinical meaning of skin data samples. The hyperspectral image data will be transformed to a more separable space using Fisher’s discriminant method. A maximum likelihood estimation algorithm will be used to cluster the image data samples to one of the skin classes with clinical meaning. During validation with real data samples, the model will take into consideration other properties such as tumor border, texture and shape features computed from the image processing algorithms. The statistical model will provide a non-invasive method for diagnosis which will both be a lower cost and safe procedure than regular biopsy methods.The initial work of this project will consist in developing the experimental setup and implementation of the modeling and image processing methods.

Phone: (787) 834 3165
Fax: (787) 834-3165
e-mail: mbrs[AT]uprm[dot]edu
University of Puerto Rico at Mayagüez
Research and Development Center
Main Building, Office 210
Mayaguez, PR