Performance of filters for noise reduction in maxillary alveolar bone imaging

Film-grain noise degrades image detail, hinders detection of subtle radiographic bone changes, and could thwart attempts to use dental radiographs of alveolar bone to detect osteoporosis. The purpose of this investigation was to quantify and compare the performance of various 1- and 2-D spatial and frequency domain filters in suppressing this noise. Estimates of noise-free bone profiles (scan lines) from each of five maxillary interdental areas were made by superimposing and averaging 16 identically exposed and digitized radiographs. The average mean absolute error and mean-squared error between the 80 initially noisy images and their respective noise-free profiles were calculated to provide an estimate of initial noise. Filter performance was measured as the change in these values after filtering the noisy images. Frequency domain analysis revealed that bone signal power spectra dominated at frequencies less than 23 cycles/mm and that some form of low-pass filtering would be applicable. The 2-D Butterworth lowpass filter provided the best performance, removing 57% of the film-grain noise when measured by mean absolute error, and over 80% when measured by mean-squared error. Surprisingly, the Lee, L _p mean, geometric mean, binomial, median, and simple neighborhood averaging filters offered comparable levels of performance.