Harvard Catalyst Profiles

Contact, publication, and social network information about Harvard faculty and fellows.

Clinical Image Retrieval: User needs assessment, toolbox development &evaluation


Advances in digital imaging technologies have led to a substantial growth in the number of digital images being created and stored in hospitals, medical systems, and on the Internet in recent years. Effective medical image retrieval systems can play an important role in teaching, research, diagnosis and treatment. Images were historically retrieved using text-based methods. The quality of annotations associated with images can reduce the effectiveness of text-based image retrieval. Despite recent advances, purely content- based image retrieval techniques lag significantly behind their textual counterparts in their ability to capture the semantic essence of the user's query. Preliminary research suggests that a more promising approach is to adaptively combine these complementary techniques to suit the user and their information needs. However, for these approaches to succeed, the researcher needs to enhance her computational skills in addition to acquiring a comprehensive understanding of the relevant clinical domain. This Pathway to Independence (K99/R00) grant application describes a training and career development plan that will allow the candidate, an NLM postdoctoral fellow in Medical Informatics at Oregon Health &Science University to achieve these objectives. The training component will be carried out under the mentorship of Dr. W. Hersh with Dr. Gorman (user studies). Dr. Fuss (radiation medicine) and Dr. Erdogmus (machine learning) providing additional mentoring in their areas of expertise.

The long-term goal of this Pathway to Independence (K99/R00) project is to improve visual information retrieval by better understanding user needs and proposing adaptive methodologies for multimodal image retrieval that will close the semantic gap. During the award period, activities will be focused on the following specific aims: (1) Understand the image retrieval needs of novice and expert users in radiation oncology and develop gold standards for evaluation;(2) Develop algorithms for semantic, multimodal image retrieval;(3) Perform user based evaluation of adaptive image retrieval in radiation oncology;(4) Extend the techniques developed to create a multimodal image retrieval system in pathology


Funded by the NIH National Center for Advancing Translational Sciences through its Clinical and Translational Science Awards Program, grant number UL1TR002541.