Active learning framework with iterative clustering for bioimage classification

Abstract

Advances in imaging systems have yielded a flood of images into the research field. A semi-automated facility can reduce the laborious task of classifying this large number of images. Here we report the development of a novel framework, CARTA (Clustering-Aided Rapid Training Agent), applicable to bioimage classification that facilitates annotation and selection of features. CARTA comprises an active learning algorithm combined with a genetic algorithm and self-organizing map. The framework provides an easy and interactive annotation method and accurate classification. The CARTA framework enables classification of subcellular localization, mitotic phases and discrimination of apoptosis in images of plant and human cells with an accuracy level greater than or equal to annotators. CARTA can be applied to classification of magnetic resonance imaging of cancer cells or multicolour time-course images after surgery. Furthermore, CARTA can support development of customized features for classification, high-throughput phenotyping and application of various classification schemes dependent on the user's purpose.

CARTA-based semi-automatic detection of stomatal regions on an <i>Arabidopsis</i> cotyledon surface

“…CARTA is a freely available tool for the classification of bioimages (Kutsuna et al 2012). We have demonstrated its application to the semi-automatic detection of biologically interesting regions, together with randomly located ROIs.…”

“…However, these approaches need enormous human effort to prepare a huge number of annotated images as training data. We have recently developed an active learning framework called ClusteringAided Rapid Training Agent (CARTA), which combined with a genetic algorithm and self-organizing map (SOM) clustering can accomplish labor-saving and high-accuracy classification of bioimages of various types including fluorescence microscope and magnetic resonance images (Kutsuna et al 2012). We can annotate an image two times faster than manual imageby-image visible inspection with CARTA while maintaining a high level of accuracy (Kutsuna et al 2012).…”

“…We have recently developed an active learning framework called ClusteringAided Rapid Training Agent (CARTA), which combined with a genetic algorithm and self-organizing map (SOM) clustering can accomplish labor-saving and high-accuracy classification of bioimages of various types including fluorescence microscope and magnetic resonance images (Kutsuna et al 2012). We can annotate an image two times faster than manual imageby-image visible inspection with CARTA while maintaining a high level of accuracy (Kutsuna et al 2012). Focusing on these advantages, we are now trying to apply the CARTA framework to detect biological objects from wide-field transmission electron microscopic images as discussed in the recent symposium cosupported by the Japanese Society of Plant Morphology and the Non-profit Organization for Integrated Imaging Research Support at the 77 th annual meeting of the Botanical Society of Japan in September 2013.…”

“…The parameters, ROI number and size, are very important for final detection accuracy, so they should be tuned appropriately. Next, SOM clustering was performed based on our original image feature set (Kbi296 features, Kutsuna et al 2012). In this SOM clustering, all ROIs were located based on their similarities using Kbi296 features (Figure 2).…”

Bioimage Informatics for Big Data

“…Efforts have been made to use machine learning to help solve these problems. Application examples include screening of gene expression patterns (Long et al 2009;Zhou and Peng 2011) and many others (e.g., Jones et al 2009;Kutsuna et al 2012). Interesting machine learning toolboxes for bioimage informatics have also been built, such as WND-CHARM (Orlov et al 2008) and ilastik (Sommer et al 2011).…”