Effects of digital image processing on the detection of simulated lesion in chest radiographs: an experimental study

ABSTRACT

The receiver operating characteristic (ROC) curve analysis was done to evaluate the effects of digital image processing techniques on the detectability of simulated pulmonary nodules and cysts in chest radiographs. Image processing techniques used were the contrast inversion, unsharp masking, and filtered equalization method. Simulated pulmonary nodules and cysts were generated by the computed program using Turbo-C language for personal computer. Each processed and unprocessed image sets of sixty images with pulmonary nodules and cysts and one hundred and twenty normal images were analyzed by three board-certified radiologists and four senior residents training in diagnostic radiology by five rating category. Area under the ROC curve was calculated using the trapezoidal rule. Mean area under the ROC curve of seven radiologists of the unprocessed image was 0.6360±0.0583, contrast inversion was 0.5660±0.0478, unsharp masking was 0.7534±0.0355 and filtered equalization was 0.6915±0.0472. Unsharp masking(p<0.01) and filtered equalization image(p<0.05) was statistically significant than unprocessed image. Digital chest radiographs processed by the unsharp masking method significantly increased the detectability of simulated pulmonary nodules and cysts overlapped with mediastinum and pulmonary cysts in free lung fields.