Automated image analysis systems to quantify physical and behavioral attributes of biological entities

Kalwa, Upender
Journal Title
Journal ISSN
Volume Title
Source URI
Research Projects
Organizational Units
Journal Issue

All life forms in nature have physical and behavioral attributes which help them survive and thrive in their environment. Technologies, both within the areas of hardware systems and data processing algorithms, have been developed to extract relevant information about these attributes. Understanding the complex interplay of physical and behavioral attributes is proving important towards identifying the phenotypic traits displayed by organisms. This thesis attempts to leverage the unique advantages of portable/mobile hardware systems and data processing algorithms for applications in three areas of bioengineering: skin cancer diagnostics, plant parasitic nematology, and neglected tropical disease.

Chapter 1 discusses the challenges in developing image processing systems that meet the requirements of low cost, portability, high-throughput, and accuracy. The research motivation is inspired by these challenges within the areas of bioengineering that are still elusive to the technological advancements in hardware electronics and data processing algorithms. A literature review is provided on existing image analysis systems that highlight the limitations of current methods and provide scope for improvement.

Chapter 2 is related to the area of skin cancer diagnostics where a novel smartphone-based method is presented for the early detection of melanoma in the comfort of a home setting. A smartphone application is developed along with imaging accessories to capture images of skin lesions and classify them as benign or cancerous. Information is extracted about the physical attributes of a skin lesion such as asymmetry, border irregularity, number of colors, and diameter. Machine learning is employed to train the smartphone application using both dermoscopic and digital lesion images.

Chapter 3 is related to the area of plant parasitic nematology where automated methods are presented to provide the nematode egg count from soil samples. A new lensless imaging system is built to record holographic videos of soil particles flowing through microscale flow assays. Software algorithms are written to automatically identify the nematode eggs from low resolution holographic videos or images captured from a scanner. Deep learning algorithm was incorporated to improve the learning process and train the software model.

Chapter 4 is related to the area of neglected tropical diseases where new worm tracking systems have been developed to characterize the phenotypic traits of Brugia malayi adult male worms and their microfilaria. The worm tracking algorithm recognizes behavioral attributes of these parasites by extracting a number of features related to their movement and body posture. An imaging platform is optimized to capture high-resolution videos with appropriate field of view of B. malayi. The relevance of each behavioral feature was evaluated through drug screening using three common antifilarial compounds.

The abovementioned image analysis systems provide unique advantages to the current experimental methods. For example, the smartphone-based software application is a low-cost alternative to skin cancer diagnostics compared to standard dermoscopy available in skin clinics. The lensless imaging system is a low-cost and high-throughput alternative for obtaining egg count densities of plant parasitic nematodes compared with visual counting under a microscope by trained personnel. The B. malayi worm tracking system provides an alternative to available C. elegans tracking software with options to extract multiple parameters related to its body skeleton and posture.

Cancer diagnostics, Computer vision, Deep learning, Image processing, Lensless imaging, Nematode tracking