2016 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) 978-1-5090-1610-5/16/$31.00 ©2016 IEEE 479 Detection of Fungal Spores in 3D Microscopy Images of Macroscopic Areas of Host Tissue Abhishek Kolagunda * , Randall Wisser † , Timothy Chaya ‡ , Jeffrey Caplan ‡ and Chandra Kambhamettu * * Computer and Information Sciences, † Plant and Soil Sciences, ‡ Delaware Biotechnology Institute University of Delaware, Newark, Delaware 19716 Abstract—The measurement of variation in characteristics of an organism is referred to as phenotyping, and using image data to extract phenotypes is a rapidly developing area in biological research. For studying host-pathogen interactions, 3D microscopy data can provide useful information about mechanisms of infec- tion and defense. Performing research on a fungal pathogen of a plant, we recently developed methods to image and combine multiple fields of view of microscopy data across a macroscopic scale. This study was focused on using macroscopic microscopy data to digitally extract the top epidermal cell layer of plant leaves and to count the number of fungal spores on the epidermis. This was achieved using an active surface approach to estimate the 3D position of the epidermis and a shape-template matching approach to detect spores. A compact shape representation is proposed to model spore shapes and generate candidate templates for detecting spores. Our experiments show results that indicate strong promise for the proposed approach in studying plant- fungal interactions. Index Terms—Spore Detection, Confocal Microscopy, Macro- scopic Microscopy. I. I NTRODUCTION AND RELATED WORKS Object detection is an extensively studied topic in the field of computer vision and image processing, which involves ex- tracting invariant features of the object that capture appearance and structure, learning to classify the features as belonging to the object or not, and searching through the target image to find regions whose extracted features match that of the object. Object detection is useful for a range of applications including robotics, surveillance, human computer interactions and medical image analysis. Recently, 3D fluorescence confocal microscopy was used to image plant leaves [15]. This approach was extended to image contiguous, millimeter-scale areas of leaves infected with fungi using 3D confocal microscopy with fluorescent markers for detection [17]. These data provide the opportunity to examine aspects of plant-fungal interactions, such as the number of spores on the surface of a leaf and those that germinate and penetrate into the plant tissue. Locating and quantifying fungal spores on inoculated leaves is an important part of understand- ing resistance to fungal growth and is the first step toward quantifying other features such as germination and infection. However, detecting spores on plant leaves is a challenging task due to variability in appearance and shape of spores, the size of the image (i.e. for macroscopic micrpscopy data [17]), the relatively tiny area spores occupy, other structures of the leaf that exhibit similarity to spores, and imperfections/noise in biological specimens. In this study, we sought to develop an object detection method for the analysis of fungal spores on plant leaves. Toshev et al. [3] and Belongie et al. [4] proposed a shape matching based object detection approach using global shape representations which requires initial segmentation to form hypotheses for the target object. Perner et al. [1] proposed a case-based object recognition approach to detect fungal spores in microscopy images. They generated prototype edge orientation templates from manually marked spore shapes that were aligned and clustered. The shape of a fungal spore was approximated by a best fitting polygon. The approach uses shape templates from specific cases which might not account for variation in the shape and could require a large number of templates for matching. Bai et al. [2] proposed a shape matching approach to detect objects using a single template. They represent the shape as a shape-band that models variation in the shape within a bandwidth of its contour. Building upon these approaches, the method described here generates weighted-boundary/blob templates by sampling can- didates from the distribution of spore shapes. Shapes represented as point distribution models and poly- gons have been used to generate statistical models of shapes and their variation [8]. However, such models are not com- pact and do not have implicit or explicit forms. Piece-wise functions such as radial basis interpolation functions [6], [7] have been used to implicitly model complex shapes, but they usually require a large number of parameters. Geometric shape models have both implicit and explicit forms [9]–[11] that can compactly model a large class of shapes. In our work, the shape of the fungal spore is modeled by a curved ellipse, and the distribution of the shape parameters are learned from manually marked spores. Candidate spore shapes (non- specific) are sampled from parameter space of this distri- bution. The shape model is used to generate spore-shaped- blob templates and weighted-boundary-orientation templates for matching and detecting spores. II. METHODS The method used for detecting fungal spores can be outlined as follows. First, because spores occur on the surface of the leaf (epidermal cell layer), the location of the leaf surface within the 3D image data is estimated. The extracted leaf surface image is smoothed and the contrast is enhanced. Since we employ a shape matching approach, the next step involves generating a boundary/edge map to be matched with spore