Premium
Lesion detection in radiologic images using an autoassociative paradigm: Preliminary results
Author(s) -
Raff Ulrich,
Newman Francis D.
Publication year - 1990
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.596449
Subject(s) - medical imaging , computer science , artificial intelligence , image processing , computer vision , radiology , lesion , medical physics , pattern recognition (psychology) , medicine , image (mathematics) , pathology
An area of artificial intelligence that has gained recent attention is the neural network approach to pattern recognition and classification. The use of neural networks in radiologic lesion detection is explored by employing what is known in the literature as the “novelty filter.” This filter uses a linear algebraic model, whereupon in neural network terms, images of normal patterns become “training vectors” and are stored as columns of a matrix. An image of an abnormal pattern is introduced and the abnormality or the “novelty” is extracted. A noniterative technique has been applied. In a preliminary experiment, autoassociative recall was tested using alphabetic characters as training vectors. The second experiment used sections of transverse magnetic resonance (MR) images (TR=3000 ms, TE=40 ms) of normal patients as the training vectors. A section of a transverse MR brain image with multiple sclerosis lesions was introduced to the filter and the abnormalities were extracted. In conclusion, a neural network based lesion detector may have great promise in medical pattern recognition.