Computer Aided Diagnosis of Glandular Tumours for Internet-Based Telemedicine
Toshiyuki Tanaka · InTech eBooks · 2011
In developing countries and isolated islands, lack of medical staff is a major problem. When a person feels unwell, he or she usually attends hospital and is assessed by a clinician and given a provisional diagnosis. When the clinician suspects cancer, a pathological diagnosis is required. This requires a pathologist, but pathologists are badly off and scarce in many countries. Since pathological diagnosis represents a subjective assessment by each pathologist, several pathologists must examine the same specimen and collate the results for accurate diagnosis, and each pathological diagnosis accordingly requires a substantial amount of time. A diagnosis support system using objective and quantitative approaches would thus be extremely useful. Moreover, the use of the Internet for diagnosis is expected to alleviate the problem of access to medical professionals in developing countries and isolated islands. Glandular cancers, such as those of the stomach, colon, prostate, and breast, are common, and we have therefore focused on these cancers in developing a computer aided diagnosis system using the Internet. In the field of medical image engineering, automatic and quantitative diagnosis systems have previously been studied by many researchers. Most such research has aimed at the development of diagnosis support systems addressing the lack of pathologists. A representative system is the pre-screening system for uterine cytodiagnosis that is currently in use. The pathologist must still perform many image processing steps to use these diagnosis systems. In research for tumour imaging, human input is still required for detection of the region of breast cancer, for identifying the glomeruli in the kidney; and so on. Although the types of tumour involved increase year by year, most research centres on tumours of the lung, colon, prostate, and ovaries, with few studies examining gastric tumours. Well-differentiated glandular cancer and adenoma differ from each other, and the classification of these tumours has recently been discussed by pathologists. However, no adequate automatic diagnosis systems for glandular tumours are available for such studies as described above, and pathologists have been eagerly awaiting a system for glandular tumours. Morphological classification of colorectal microscopic images has been reported as a recent diagnosis support system. This method allows features of tumours to be obtained from the whole image, but has a low rate of correct classification for images with a relatively large background region. With the aim of formulating an Internet-based diagnosis system,