Date of Award

Winter 2016

Project Type


Program or Major

Electrical and Computer Engineering

Degree Name

Master of Science

First Advisor

Richard A Messner

Second Advisor

John R LaCourse

Third Advisor

Wayne J Smith


Disseminated neoplasia is a type of cancer that is prevalent in marine bivalves. A group of biologists at the University of New Hampshire are frequently tasked with estimating the proliferation of this cancer in soft-shell clam hemocytes using unstained samples and a bright-field microscope without the use of any real means of quantization. Instead, their measurement is a purely visual analysis of confluency where varying individual experience amongst researchers, no usage of stains, cell clustering and the general nature of the microscopy environment make it exceedingly difficult to perform this task with consistent accuracy. This thesis details the application of image processing and machine learning to streamline this process in an effort to reduce human error and provide researchers with more reliable identification of cancerous hemocytes from photomicrographs. This work contains analysis of noise in input images, its sources and the preprocessing methods used for mitigation. A segmentation algorithm capable of isolating individual cells from each image is also explained. This is followed by an assessment of structural, morphometric and densiometric features extracted from regions of interest. Finally, discussion of the applicability of these feature vectors to train various classifier models is presented. The training dataset used for this work was extremely skewed due to an over abundance of healthy cells relative to cancerous ones leading to additional consideration of a secondary dataset modified to alleviate skew. The differences between the most successful classification methods are described for both datasets, including their success in locating new cells not included in the training data. While not entirely free of errors (~95\% theoretical accuracy), this algorithm successfully provides researchers with an improved means for identifying cancerous hemocytes.