Effect of digital noise filters on diagnostic radiographs for the diagnosis of experimental root fractures.

This article discusses basic concepts of noise suppression for digital radiographs, and shows how false application of algorithms available can lead to a loss of information. Noise suppression realized with low pass convolution filters are designed to decrease strong grey level changes of neighboring pixels. Sixty-two extracted fully intact premolars were selected and treated usingthe crown-down technique to prepare them for experimental fractures. A total of 358 radiographs were taken, of which 312 images were selected with clearly visible root fractures. The selected radiographs were stored as uncompressed images and manipulated with 8 different convolution filters. A total of 2496 raw and filtered images were inspected for visible root fractures. Only the arithmetic mean filter with 5 x 5 kernel size led to a deletion of the visible fracture line in 33.3% of the inspected cases. The other filters proved to be useful for noise reduction without loss of any necessary diagnostic information. None of the filters improved the diagnostic outcome in the 46 remaining radiographs which did not show clearly visible root fractures before application of the filters.