SBIR-STTR Award

The overall goal is to develop tools to assist physicians in using imaging devices to reliably detect malignant melanoma in its early stages. The proposed effort has three aims: 1) to determine feasibility of detecting early melanoma through extraction of information - not otherwise available to the physician - by wavelet transformation of multispectral images of cutaneous lesions; 2) to combine diagnostic parameters extracted from such wavelet representations with other parameters obtained from untransformed grey-level images; and 3) to develop wavelet-based compression methods applicable to multispectral images for telemedicine - and suitable for both human and machine vision. The first two goals are for Phase 1. If either or both of these goals are met, the third one would be pursued in Phase 2. PROPOSED COMMERCIAL APPLICATION: Potentially standard PC-based wavelet transformation software, both for lesion image representation and for image compression. Would be applicable both to conventional and to dermoscopic images obtained with a film camera or with a CCD- based multispectral imaging system.

Thesaurus Terms:
computer assisted diagnosis, computer system design /evaluation, digital imaging, image processing, melanoma, neoplasm /cancer diagnosis computer program /software, early diagnosis, method development data collection, histopathology, human data, human subjectNational Cancer Institute (NCI)

A practical, digital dermoscopic imaging system will be developed for the reliable, non-invasive, early diagnosis of melanoma. This system will provide an objective and user-friendly tool to assist health care providers in diagnosing melanoma. Phase 1 produced two very important results: 1. The feasibility of a new paradigm for reliable, automatic diagnosis of early melanoma was demonstrated. It combines wavelet-based multi-scale statistical parameters, that quantify textures of dermoscopic images, with other, non-wavelet parameters reported by us in the literature. Significantly improved differentiation between early melanoma and its benign simulants is achieved thereby. 2. The feasibility of using well calibrated, multispectral, dermoscopic lesion images for automated lesion segmentation, feature extraction and classification was established. The specific aims of Phase 2 are: (l) Further develop the image database for training and more comprehensive testing of our diagnostic methods. (2) Analyze parametrically the dependence of performance of the proposed lesion classification methods on the spectral illumination bands, spatial resolution, and dynamic range of the imaging system. (3) Design and build six commercial prototypes. (4) Clinically test these prototypes at four major clinical centers. PROPOSED COMMERCIAL APPLICATIONS: This imaging system could become standard instrumentation in the offices of dermatologists and health care delivery facilities for melanoma screening, monitoring of suspicious pigmented lesions, and as an aid in their diagnosis. Also, the database developed during Phase 2, and beyond, will be an important medical resource for the health-care community.

Thesaurus Terms:
computer assisted diagnosis, computer system design /evaluation, digital imaging, image processing, melanoma, neoplasm /cancer diagnosis artificial intelligence, computer program /software, early diagnosis, method development, noninvasive diagnosis, training bioimaging /biomedical imaging, clinical research, data collection, histopathology, human data, human subjectNATIONAL CANCER INSTITUTE