Using active learning to annotate microscope images of parasite eggs

Abstract

Microscopic analysis forms an integral part of many scientific studies. It is a task which requires great expertise and care. However, it can often be an extremely repetitive and labourious task. In some cases many hundreds of slides may need to be analysed, a process that will require each slide to be meticulously examined. Machine vision tools could be used to help assist in just such repetitive and tedious tasks. However, many machine vision solutions involve a lengthy data acquisition phase and in many cases result in systems that are highly specialised and not easily adaptable. In this paper, we describe a framework that applies flexible machine vision techniques to microscope analysis and utilises active learning to help overcome the data acquisition and adaptability problems. In particular we investigate the potential of various aspects of our proposed framework on a particular real world microscopic task, the recognition of parasite eggs.

A survey for the applications of content-based microscopic image analysis in microorganism classification domains

“…Fig. 13 The active learning process: active learning is a form of supervised learning where the learner queries an oracle to get class information for unlabelled training examples (Nugent et al 2006) In Nugent et al (2006), a multi-class classification method is developed to find different parasite eggs, where colour features, an active learning framework and a ranking classifier are used. In the experiment, three MM types (59 images) are tested, and an overall accuracy higher than 95% is achieved.…”