A.G. Todman

Colour texture analysis using co-occurrence matrices for classification of colon cancer images

Worldwide, colorectal cancer is the third most common malignant neoplasm. Automated classification of cytological images of colon tissue samples has been investigated, but diagnosis in all cases still requires human judgement. With the large numbers of cases of colon cancer each year, the workload placed on pathologists is immense. Texture is a powerful discriminating metric and the use of grey-level texture for classification of colon images has been extensively researched. One common technique is the extraction of texture metrics from grey-level co-occurrence matrices. However, using grey-scale images discards information contained in the differences of hue and saturation that may provide further classification information. We present the findings of an investigation of the discriminating ability of colour texture using co-occurrence matrices. Comparisons are made between grey-scale and colour texture analysis. Using statistical analysis, we show that classification using colour texture offers an improvement over classification based solely on grey-level texture.

Breast self examination training through the use of multimedia: a benchmark multimedia development methodology for biomedical applications

This paper discusses the development of a set of specifications for a development methodology for educational multimedia, specifically for the development of a prototype educational multimedia application for Breast Self Examination (BSE) training. In the paper, the need for BSE training is examined and BSE is briefly presented. Our focus, however, is to show how the development of a multimedia authoring methodology could provide an effective solution to BSE and other biomedical training needs. A method for setting the overall specifications for a multimedia development methodology for biomedical applications, in general, and a BSE training multimedia application, in particular, is presented.

Multiresolution colour texture analysis for classifying colon cancer images

Co-occurrence matrices are commonly used to extract fine texture information from images, and have been found to be a useful tool for measuring dysplasia in histological images of the colon. Pathologists, however, measure dysplasia in tissue samples at structural as well as cytological levels. We present our findings after investigating modifications to the cooccurrence matrix technique to measure this low frequency colour texture information for the classification of colon cancer images.

Breast self examination training through the use of multimedia: a prototype multimedia application

This paper presents the final version of a prototype multimedia application for training women on performing breast self examination (BSE) in the right way. The application was developed for the purposes of an evaluation study on the effectiveness of three different BSE training media: leaflets, videos and multimedia applications. Since the British National Health Service (NHS) does not offer BSE training at the moment in the form of a multimedia application, a prototype had to be developed and compared against the existing material of leaflets and videos. The paper presents a novel video playback approach, the main differences between the BSE multimedia application and other traditional training approaches already in use by the NHS at the moment, the benefits of using multimedia in biomedical training and education and the final applications structure along with the evaluation criteria for the applications effectiveness and the comparison methodology that will be used to compare its effectiveness against leaflets and videos.

Orientational coherence metrics: classification of colonic cancer images based on human form perception

A number of quantitative techniques for the analysis of histopathological images used in the diagnosis of colonic cancer have been researched in the literature. While these methods have led to significant advances in the development of automated techniques, manual processing by a clinical expert remains the standard against which results are assessed. Here we explicitly attempt to build our understanding of human form perception, implemented at a neural level, into metrics that give us a measure of the extent to which the structure in an image displays a coherent orientational specificity. We describe the derivation of these metrics and consider their application to typical images of normal colon, dysplastic specimens, and moderately and poorly differentiated adenocarcinoma of colon. Preliminary results evaluating the effectiveness of the total activation ratio and the orthogonal activation ratio suggest that they are capable of separating images of normal colon tissue from those of moderately and poorly differentiated adenocarcinoma of colon.

Learning histopathological microscopy

Histopathological tissue analysis by microscopy is a process that is subjective, prone to inter- and intra-observer variation. This, along with the problems associated with verbalising visual elements of the diagnostic process, make learning the skill quite difficult. Training is long and largely relies on an “apprentice” model, where trainees learn the skill by witnessing an expert at work. Here we present the first findings of a longitudinal study of a group of histopathology trainees. By monitoring the progress of the trainees, we hope to be able to provide information that will improve training and assessment. In this paper we discuss the results of early data collection and analysis, from a web-based study of trainee classification accuracy and classification time.

A prototype multimedia application for breast self examination training

This paper discusses the development of a prototype multimedia application for training women to perform Breast Self Examination (BSE) correctly. The paper presents the main differences between the BSE multimedia application and the other, traditional training approaches currently in use by the UK National Health Service (NHS) at the moment. The benefits of using multimedia in biomedical training and education are discussed. An overview of the applications structure and design is presented along with the evaluation criteria for the applications effectiveness.

IRiS: an interactive reality system for breast self-examination training

This paper presents a novel interactive reality video playback approach developed for biomedical training purposes, and tested on a prototype breast self-examination (BSE) multimedia training application. The system is developed in order to improve on existing video playback approaches as used in multimedia applications by providing control over not only time, as in conventional video playback, but also space. The benefits of interactive reality video playback are presented and the approach is compared with other similar approaches, such as QuickTime and iPIX. The design, development, final implementation, testing and evaluation plan of the IRiS system are presented. The paper also discusses future plans and the use of the system in other biomedical training scenarios.

A real-time hand tracking and multimedia-based system for breast self-examination: the simplified 3D hand model

Breast self-examination (BSE) is a non-invasive, self-administered and simple screening procedure for detecting breast cancer at an early stage. This procedure can be performed in private and at any time. A variety of leaflets and websites exist, which attempt to train women on how to perform BSE. There are also some learning systems consisting of videos and audio cassettes. However, there are no fully interactive systems in existence and no real-time feedback is given to a user on whether she is correctly performing the procedure. We aim to develop an intelligent interactive multimedia system incorporating pattern recognition and machine vision techniques, and provide real-time feedback to assist and guide women to perform BSE accurately. Using her hand in a specific configuration to conduct palpation of the breasts is the basic means available for a woman to perform BSE. However, a human hand is highly articulated and deformable with 27 degree-of-freedom parameters according to its anatomy. Recognising hand gestures and postures is a challenging task that has been studied in many areas and applications. In this paper, the simplified three-dimensional (3D) hand model is presented, which has only 8 degree-of-freedom parameters and is especially adapted for use with the breast self-examination system. This model will be a potentially effective simulation and tracking tool that will contribute to BSE learning and thus to the development of an intelligent fully interactive BSE system.

Hand motion segmentation against skin colour background in breast awareness applications

Skin colour modelling and classification play significant roles in face and hand detection, recognition and tracking. A hand is an essential tool used in breast self-examination, which needs to be detected and analysed during the process of breast palpation. However, the background of a womans moving hand is her breast that has the same or similar colour as the hand. Additionally, colour images recorded by a web camera are strongly affected by the lighting or brightness conditions. Hence, it is a challenging task to segment and track the hand against the breast without utilising any artificial markers, such as coloured nail polish. In this paper, a two-dimensional Gaussian skin colour model is employed in a particular way to identify a breast but not a hand. First, an input image is transformed to YCbCr colour space, which is less sensitive to the lighting conditions and more tolerant of skin tone. The breast, thus detected by the Gaussian skin model, is used as the baseline or framework for the hand motion. Secondly, motion cues are used to segment the hand motion against the detected baseline. Desired segmentation results have been achieved and the robustness of this algorithm is demonstrated in this paper.