Abstract:
The objective of this thesis is to evaluate the diagnostic loss in compressed medical images using computer simulation. Compressing medical images is a necessity due to the cost of the storage medium as well as the low bandwidth available for telemedicine procedures. Experimental studies conducted conventionally for this purpose use a set of real images for which a consensus is reached by a team of medical imaging specialists, on the presence or absence of a lesion. Then ROC (Receiver Operating Characteristic) curve analysis is carried out in order to determine the e ect of compression at di erent ratios in terms of lesion detectability. The area under the curve (AUC) equals one when lesions can be detected perfectly well. If they can not be detected the area under the curve (AUC) equals 0.5 and this means that it is not better than arbitrary guessing. These experiments should be conducted by using many images and observers if it will be statistically signi cant. Therefore it is time consuming and expensive. Furthermore, this method has serious drawbacks since it does not include any analysis for small subtle lesions and is impossible to compare the errors due to other factors such as variation in equipment and data acquisition protocols. This thesis has the objective of eliminating these drawbacks by using a computer simulation of the entire imaging chain that includes the organ, the imaging equipment and the human observer. A Monte Carlo simulation package (SIMIND) has been used to simulate the image formation process for a gamma camera acquiring data from a breast containing a lesion. The obtained images are then compressed using the JPEG and JPEG 2000 algorithms at di erent compression ratios. Lesion detectability is then assessed by using a mathematical observer model named the channelized hotelling observer. Image quality is also assessed using quantitative image quality metrics. The results showed that diagnostic loss occurs at all compression ratios for subtle lesions but this loss may be comparable to other losses such as the ones due to variation in equipment and data acquisition protocols. Eventually, the decision of which compression rate to adopt will not be di erent than any other engineering tradeo decision made for balancing cost and performance. This is in contrast with experimental studies that determine the ideal compression ratio based on evident lesions only and therefore presents an alternative methodology.
