Image Analysis of Histological Features in Molar Pregnancies

Molar pregnancy (also known as hydatidiform mole, hydatid mole, gestational trophoblastic disease) represents forms of abnormal conception caused by defective fertilisation resulting in excess expression of paternal genes in placental tissue. There are two forms of hydatidiform mole: complete (diploid androgenetic) and partial (paternal triploid), the distinction between which is important for determining appropriate prognosis and management of patients. Both complete and partial hydatidiform moles are associated with increased risk of development of malignant gestational trophoblastic tumours, the risk being much greater for complete hydatidiform moles. Whilst in most cases the diagnosis of these moles can be reliably achieved on morphological histological assessment, these represent a continuing diagnostic problem for histopathologists since in early pregnancy complete hydatidiform moles, partial hydatidiform moles and non-molar hydropic miscarriages may be difficult to distinguish.

In this paper, we propose a computational image analysis approach guided by the knowledge of expert pathologists in identifying essential distinguishing morphological criteria. The approach, which combines Fuzzy C-Means clustering with hue, saturation and value colour space, shows promising results as it is able to classify successfully the villi into appropriate regions, namely trophoblast and stroma, and extract areas of blood. However, because of the marked variations in size, shape and outline of the villi, and trophoblast proliferation, both within and between cases, the analysis shows that there is no single criteria which can reliably classify these products of conception and a combination of criteria is required.