Interactive segmentation and tracking in optical microscopic images

RECENT advances in optical microscopy have caused a data explosion as scientists can observe biological processes at a wide range of spatial and temporal resolutions. Hence, computer-aided microscopic image analysis is essential in utilizing the complete information embedded in such large volumes of data as manual analysis is extremely time consuming. Segmentation and tracking of objects of interest are some of the fundamental steps in a wide range of microscopic image analysis tasks and have been active areas of research for the past several decades. Though critical, these problems are extremely challenging not only due to ambiguities introduced by the imaging system but also due to wide variations in target objects of interest. Being the smallest units of life, cells and also nuclei have been popular choices as objects of interest. In this context, segmentation and tracking of cells and nuclei in tissue sample images is of particular importance in developmental biology and tumorigenesis since we can observe cells in their natural surroundings, which give us much more realistic experimental conditions compared to a cell culture model. However, development of such algorithms for tissue images is usually far more challenging compared to a cell culture model due to poor signal-to-noise ratio and the inherent contact between cells, making it extremely difficult to segment and/or track structures accurately. Hence, several automatic (1) and semi-automatic (2) methods have been developed to perform tissue segmentation and tracking. The ultimate goal of developing a segmentation and tracking algorithm is to automate it, so that the processing can be done without any manual intervention. This not only increases the throughput of the method enabling it to be used in a wide variety of applications where large datasets have to be analyzed in reasonable amount of time but also reduces the bias and subjectivity introduced by a manual analysis. This is also important in a clinical setup (computer-assisted diagnosis) where on one hand accuracy and reliability of the result is of utmost importance, and on the other hand timely analysis of the data is also required. However, due to aforementioned difficulties and regulatory policies, complete automation is not always possible. Hence, several interactive semi-automatic segmentation and tracking algorithms have been developed which although require some manual intervention, minimizes it as far as possible. The manual interaction is often incorporated in two ways, (I) a continuous corrective interaction with the data and intermediate segmentation and/or tracking results or (II) as a corrective post processing step after the data is segmented and tracked using a fully automatic method. There are several advantages of using a semi-automatic method. In almost all datasets, including the challenging ones where automatic methods have very little hope of producing a satisfactory result, the user can get the help of a semi-automatic method to segment and track objects. All interactive methods allow users to adjust and improve the segmentation or tracking results until they satisfy their own quality control standards. Also, similar to the automatic methods, the semiautomatic methods help in reducing the subjectivity, bias and inter-user variability introduced by a completely manual method to an extent by allowing the user to take the optimal