Confocal Imaging of Skin and the Oral Cavity
Charles DiMarzioAssociate Professor, Department of Electrical and Computer Engineering, and Milind Rajadyhaksha, Assistant Member, Dermatology Service
Northeastern University, Boston, MA, and Memorial Sloan Kettering Cancer Center, New York, NY
Background and GoalsDermatology research and practice is currently hampered by a lack of imaging devices at an appropriate microscopic scale. Clinicinas use visual inspection for most examinations, often aided by a "magnifying glass" type device known as a dermascope. If they wish to obtain higher resolution than available with such a device, the next step is typically excision followed by histology. This is true for screening as well as when verifying clean margins in lesion excision procedures. In response to this need, Dr. Rajadhyaksha and Prof. DiMarzio, along with collaborators at Lucid, Inc., in Rochester NY, have been working for many years on the development of flexible low-cost reflectance and fluorescence confocal microscopes for use in this application. Indeed Lucid has produced a series of such instruments marketed under the name Vivascope.
However images from such devices present challenges to clinicians: distinct from the traditional histological slice, which is en face in orientation, confocal images are presented as slices more-or-less parallel to the skin surface. Moreover they are low-contrast images, especially in the (non-invasive, and therefore preferred) reflectance modality. CIBC, along with Northeastern University researcher Prof. Jennifer Dy, have been working with Dr. Rajadhyaksha and his colleagues at Sloan Kettering to develop algorithms for semi-automatic and automatic analysis of these images. As a first goal, we have been working to detect the three-dimensional boundary between the epidermis and dermis in these data, as it is the dominant physical feature of the skin as an organ and also is the locus of most incipient oncologically-relevant events. We have had some success and are now moving into a validation phase using hand-segmentations by clinical experts as our metric. We note that this work has attracted the attention of a pharmaceutical company with interest in observing such morphological features of the skin non-invasively, and we are in the last stages of negotiating for support from them which we hope will dramatically enhance our validation efforts. However much remains to be done to move this effort into the range of clinical relevance, including study of lesions (so far we have only looked at normal skin) and development of machine learning algorithms for classifying lesions as tumor, dysplasia, normal, etc.
There is still considerable room for improvement in the area of clinical application of confocal imaging through development of novel instruments. Both Dr. Rajadhyaksha and Prof. DiMarzio are currently engaged in such efforts, but individually and collaboratively (Dr. Rajadhyaksha is currently co-supervising a Ph.D. student in Prof. DiMarzio's lab at Northeastern.) But such development is hampered by the lack of a solid computational prediction of optical transport in realisitic skin models. We propose to use CIBC technology to help our collaborators build such a model. CIBC model-building tools such as Seg3D and BioMesh3D will be used to extract relevant structure from confocal images and construct a computational mesh. PDE solvers such as finite difference and finite elements in SCIRun will be adapted to solve the diffusion equation typically used to model optical transport in highly-scattering media such as tissue. Boundary conditions relevant to a particular proposed imaging device will be developed and implemented as SCIRun modules by our collaborators. The result will be a significant contribution to the ability of microscope innovators to rapidly prototype a new or proposed device in silico to test feasibility, obtain bounds on potentially achievable resolution, and guide further device design. As a consequence we expect cost of instrument development to be reduced and timelines to become significantly shorter.
PublicationsS. Kurugol, J. Dy, M. Rajadhyaksha, and D.H. Brooks. "Detection of the dermis/epidermis boundary in reflectance confocal images using multi-scale classifier with adaptive texture features". In Proceedings of the International Symposium on Biomedical Imaging (ISBI), Paris France, May 2008. IEEE.
S. Kurugol, J. Dy, M. Rajadhyaksha, and D.H. Brooks. "Localizing the dermis/epidermis boundary in reflectance confocal microscopy images with a hybrid classification algorithm". In accepted for International Symposium on Biomedical Imaging (ISBI), Boston MA, June 2009. IEEE.
H. Sierra, C.A. DiMarzio, and D.H. Brooks. "A product-of-convolutions model for three-dimensional microscopy, comparison to born and rytov models". In SPIE Frontiers in Optics 2006/Laser Science XXII, OSA Technical Digest (CD) paper FMI2, Rochester, NY, USA, 2006.
H. Sierra-Gil, C.A. DiMarzio, and D.H. Brooks. "Modeling dic microscope images of thick objects using a product-of-convolutions approach". In OSA Biomedical Optics Topical Meeting, page abstract. OSA, 2009.
H. Sierra, C.A. DiMarzio, and D.H. Brooks. "Modeling phase microscopy of transparent 3d objects: A product of convolutions approach". JOSA A, 26(5):1268–1276, May 2009.