Applications of image processing and machine learning techniques in wildlife monitoring and cancer cell characterization

Journal Title
Journal ISSN
Volume Title

This dissertation presents the application of state-of-art image classification and object detection techniques to two projects in the areas of wildlife monitoring and cancer cell characterization. The following presents a summary of the problem statement, solution approach, and major contributions for each of the two projects. In the wildlife monitoring project, an automated vision system is proposed for animal detection in trail-camera images taken from a field under the administration of the Texas Parks and Wildlife Department. As traditional wildlife counting techniques are intrusive and labor-intensive to conduct, trail-camera imaging is a comparatively non-intrusive method for capturing wildlife activity. However, given the large volume of images produced from trail cameras, manual analysis of the images remains time-consuming and inefficient. For this purpose, a two-stage deep convolutional neural network pipeline is implemented to find animal-containing images in the first stage and then process these images to detect birds in the second stage. The animal classification system classifies animal images with 93% sensitivity and 96% specificity. The bird detection system achieves better than 93% sensitivity, 92% specificity, and 68% average Intersection-over-Union rate. This dissertation also addresses post-deployment issues related to data drift for the animal classification system as image features vary with seasonal changes. This system utilizes an automatic retraining algorithm to detect data drift and to update the system when necessary. Moreover, two statistical experiments are presented to explain the prediction behavior of the animal classification system. These experiments investigate the cues that steer the system towards a particular decision. Statistical hypothesis testing demonstrates that the presence of an animal in the input image significantly contributes to the system’s decisions. The cell characterization study investigates the automatic detection and enumeration of circulating tumor cells in patient blood samples. Circulating tumor cells (CTCs) are invaluable biomarkers used in the early diagnosis and treatment of cancer. Microscopy images of isolated CTCs from patient blood samples are routinely acquired and analyzed for CTC detection and enumeration purposes. Due to the scarcity of CTCs in the patient blood sample, their manual characterization is a challenging task that involves a series of tedious cell staining and labeling procedures and laborious manual identification of the cells. This study proposes an automated detection and enumeration system to alleviate the lag in the enumeration process. The core of this system is an efficient and accurate convolutional neural network (CNN)-based model that performs label-free detection of MCF-7 breast cancer cells, as a proxy to CTCs, in brightfield images. The MCF-7 detection model achieves above 99% sensitivity and specificity. In addition, the average Intersection-over-Union rate of the proposed detector is better than 80%. For the training set generation, a fully automated workflow is presented that facilitates the efficient and easy labeling of brightfield images. Additionally, multiple experiments are designed and implemented to explore the prominent features that the CNN extracts and uses for distinguishing MCF-7 cells from white blood cells. The results of the experiments indicate that the designed CNN uses the size of the cell as the prominent distinguishing feature. Furthermore, if the size feature is eliminated, the CNN is still capable of extracting other features to distinguish MCF-7 cells, but with a 3% accuracy reduction. Finally, the robustness of the proposed MCF-7 detection model in the presence of various image intensity transformations is investigated. The results indicate that the F1-score of the detection model deteriorates by less than 0.2% in the presence of image intensity and contrast transformations. Therefore, the MCF-7 cell detection model has sufficient robustness with respect to variations in the intensity and contrast of the brightfield images.

Embargo status: Restricted until 06/2022. To request the author grant access, click on the PDF link to the left.

Machine Learning, Image Processing, Wildlife Monitoring, Cancer Cell Characterization, Deep Learning, Convolutional Neural Networks