Dermatology is in part an imaging specialty. Automated programmes based on analysis of two-dimensional images, such as those provided by dermoscopy, have achieved similar diagnostic accuracies to those of humans in pilot studies.

One area in which such technology is inefficient is that it fails to represent the three-dimensional (3D) nature of many skin lesions. In the present approach we use extremely close-up stereophotogrammetry to capture microscale 3D structure. This provides an extra modality for segmentation and classification, as well as enabling enhanced visualization.

In collaboration with Dimensional Imaging, we have extended their PSP® technology. The benefits of this approach over laser scanning are instantaneous capture with no artefacts due to subject movement; extremely dense data calculated at every image pixel; true one-to-one colour registration unaffected by any structured lighting; and that it is completely noninvasive.

We have used two Canon EOS 350D cameras in a bespoke frame, using ring flash to ensure uniform lighting, and a Macbeth colour chart is included in every shot. We can recover an extremely detailed model of the lesion from which the image colour and range data can be re-projected as a 3D model. The interpixel spacing is around 25 pixels mm­1 and depth accuracy is about 0.05 mm. Enhancement of the initial capture uses a specialized interface allowing visualization of a section of the particular region in question. It also performs colour correction in order to normalize the data across all captures.

Using the corrected data we can explore the additional depth information to help in segmenting the lesion, in combination with colour data. Analysis of the surface geometry also leads to further features that may be extracted for classification purposes (for instance, roughness in a seborrhoeic keratosis). (Images will be available on the poster in addition to images that can be rotated in real time on a laptop PC).

Corresponding author e-mail: jonathan.rees@ed.ac.uk