Columbia University

Technology Ventures

Interactive Segmentation imaging of Retinal Disorders

Technology #2613

“Lead Inventors: Theodore Smith, MD PhD., Andrew Laine, PhD, Noah Lee;

Age-Related Macular Degeneration imaging technique:
Macular diseases such as age-related macular degeneration, diabetic retinopathy, and macular dystrophy account for the overwhelming majority of blindness in the US. Longitudinal analysis of multiple types of (i.e. multimodal) retinal images is used to predict the onset, aid in the diagnosis, and track the progression of these retinal disorders. Retinal image analysis, however, requires time-consuming examination by qualified and experienced ophthalmologists. Some automated or semi-automated retinal image analysis software packages have been developed but they are derived from engineered models heavily based on assumptions about the objects of interest. A need exists for more generalized and robust segmentation solutions that require less intervention time from the clinician.

Bayesian transduction and Markov Random Field Models are used to segment and register retinal images :
This technology represents an algorithm for applying machine learning to the spatial classification and labeling of multimodal retinal images. Bayesian transduction and Markov Random Field Models are used to segment and register images while the transductive formalism provides predictive confidence of classification. Clinicians are required only to provide rough and sparse labels of objects of interests and the algorithm proceeds to accurately label the remaining unlabeled set of objects. This method can be integrated into existing retinal imaging and analysis systems for semi-automated registration and segmentation of multimodal retinal images. Image registration and spatial classification of objects of interest provides a means for longitudinal analysis of disease progression.

Applications: • Early detection and long term monitoring of macular diseases in a quantitative manner • May be used as a research tool to identify relationships between spatially classified objects and disease phenotype (e.g. identifying changes in optic nerve head morphology) • Segmentation algorithms can also be applied to other imaging modalities and organ systems

Advantages: • Software is easily incorporated into clinical ophthalmologists' workflow though an intuitive interface to quickly perform diagnostics on multimodal retinal images • Minimizes expert intervention time by requiring only sparse labeling by the clinician

Patent Status: Patent Pending

Licensing Status: Available for Licensing and Sponsored Research Support

Publications: 30th Annual IEEE EMBS Conference, Vancouver, Canada. August, 2008 pg. 2242-2245 ”