Computer-aided diagnosis by holographic image recognition
Emilio GOMEZ GONZALEZ
________
University of Seville
E.S.I. – Department of Applied Physics
Camino de los Descubrimientos s/n
41092 Seville, Spain
E-mail : egomez@cica.es
Nowadays computer-aided diagnosis is one of the most active fields of research in medical imaging. Also, social interest is deeply focused on screening programs for early detection of diseases such as lung and breast cancer. However, ranges of undetected positives up to 30% in bronchogenic carcinomas and 28% in breast malignant lesions, a substantial amount of which are evident in retrospective, have encouraged developments mainly oriented to assist radiologists in image interpretation and analysis.
On the other hand, physical holographic recognition has been largely devoted to military applications such as synthetic-aperture radar image analysis and target tracking systems using sophisticated optoelectronic equipment. Digital developments have also appeared in binary character readers and similar devices.
Presented method is thus an extension of digital holographic recognition suitable for use with complex medical images which allows for a very precise numerical evaluation of the degree of coincidence of two images. If one of them is that of a confirmed lesion malignant breast or lung nodule- this procedure yields a resulting matrix of coincidences which can be visualized using any user-friendly interface. Positive coincidences are given by local maxima of values with radial symmetry whose height, broadening of surrounding area and peak value define the degree of similarity to the employed reference.
Use of a reduced database of references has been found to achieve correct location of lesions in images of different patients in which nodules may appear with different densities and in another scale, size or orientation. Numerical accuracy achieved in peak evaluation ranges up to 10-15 far beyond an observers eye- and allows for discernment of subtle differences among neighboring pixels thus yielding true positive findings which may be partially hidden in a diffuse, complex background.
A subject of currently on-going research, this procedure has been applied to detection of malignant lesions in mammograms, computed tomography chest scans and conventional thoracic radiographs. Promising preliminary results, applicability to any type of digital image, and possibility of use in personal computers define main features of this procedure as a potential useful tool for medical image analysis and interpretation.