Medical radiograph classification by pattern recognition

Labrador retrievers can be affected by a syndrome that is characterized by ocular and skeletal dysplasia. The skeletal dysplasia takes the form of bone lesions that are confined to the appendicular skeleton. These lesions are characterized by shortened and abnormally shaped bones, and abnormal joint morphology. Elimination of the abnormal gene from the breed would require the identification of carrier animals by test mating with an affected animal. A method for reliably identifying affected puppies in resulting litters is to use the consistently expressed skeletal changes to make the diagnosis. The paper reports the research aimed at automating this diagnostic task. The procedures presented use hand digitized bone outline data obtained from a radiograph and compare this outline to known but 'fuzzy' shape characteristics of normal and abnormal bones. Experiments using this automated diagnostic method on thirty-one puppy bone radiographs yielded very good classification accuracies. >