| Περίληψη: | Recently, research studies have shown a rapid increase of several types of cancer which are attributed to many factors such as air pollution, unhealthy diet, modern living, random mutations etc. An important type of cancer among men that is clearly accused of high death rate if it is not diagnosed in time, is prostate cancer. Prostate cancer is being diagnosed after following a specific protocol which of course includes a TRUS-guided biopsy, usually after having an MRI scan. US has been undoubtedly considered as an imaging modality which can compete with MRI due to its low cost, the real-time imaging possibility, its portability, its non-radiation effects and its relatively easy user’s operation. On the other hand, its poor spatial resolution, often an MRI or CT-scan is required to make an accurate diagnosis of a disease regarding soft tissues. CEUS has come to bridge this gap, since it introduces UCAs, non-linear scatterers of the sound wave. These agents, often called MBs, are infused in the blood in a specific region to be scanned. A US transducer produces sound waves at a specific frequency. The MBs reflect the transmitted signal, since they are strong scatterers of sound waves, and the reflected signal is received by the transducer again. This procedure is based on the piezoelectric effect of certain materials. This means that when an electric field is applied to them they produce sound waves, but also the opposite. Thus, an image with high contrast can be achieved. Of course there are several methods that utilize the non-linear effect of the MBs and distinguish it from the surrounding tissues, called the contrast-specific imaging techniques. Many of them take advantage of the high mechanical index, while others use the low mechanical index, especially when continuous real-time imaging is necessary. This modality can be used for visualizing the micro-vascularity of the prostate gland in super-resolution using accurate localization-based detection procedures. Since, it is known that there is high correlation between micro-vascularity and cancer growth, it is very important to create US images which can provide the clinician with information about the vascular bed and possibly a tumor growth at early or more progressive stages. After detecting these MBs in super-resolution images, there comes the part, where the tracking procedure is taking place. Usually, a frame sequence of the contrast-enhanced images of the prostate is available and the MBs detected using localization techniques, have to be linked in sequential frames in order to create the tracks of each MB and thus the vessels in which the MBs are flowing into. Several Algorithms have been developed for dealing with the tracking problem, mainly divided in two sub-categories, the deterministic and the probabilistic. The deterministic approach is assumed to be a more conservative method for dealing with tracking and it is based on the optimization of the assignment problem like the Nearest-Neighbor algorithm, after having accurately localized the targets. The probabilistic approaches, on the other hand, have been recently introduced to deal with the tracking problems and it seems that they lead to more accurate and reliable results than the previous ones, since they are based on the prediction of the trajectory using the prior knowledge of measurements like locations or velocities, thus creating realistic probabilities of a certain MB to be linked with another MB. In this study a new tracking algorithm is introduced, the BM3D plus Rho, which utilizes the correlation between MBs in sequential frames to form links inside tracks. Of course, a thorough investigation of MB size and other characteristics is preceded. The new algorithm uses data taken from manually collected tracks in a sample of the population in order to optimize the linking procedure on a link-based level. A new detection filter is also added which leads to even more accurate localizations. During the analysis, an evaluation procedure is taking place using synthetic data produced by an algorithm, and thus a more robust comparison can be achieved. In this study, It was found that the vector angle correlation between MBs inside the same track range from 0 to 23 (angle in degrees) for the sample and dataset SRI001, which of course can be used in other datasets such as SRI010, in which the resulting maps were clearly improved even if the maximum correlation value is not know from a respective sample, like in SRI001. Resulting track number maps, velocity maps and blood flow maps show that the vessels can be clearly separated and the noisy areas have been reduced compared to the previous version of the algorithm. Thus the micro-vascularity of the prostate in both datasets with different MB size distributions and PSFs is depicted clearly, and of course enhanced correctly in the expected areas. It was found also, that the long tracks which the algorithm calculated do not locate inside the cancer area, but in the healthy regions, and this information is important for assessing also the differences between the cancer and the healthy area. The detection due to the filter showed an almost 2% improvement in the detection part. The tracking was improved by a significant percent since the vessels in the maps are very well structured and also the areas which contain already confirmed vessels have greater amplitude than before. The mean link per track has risen about 6% while the total tracks and links formed have risen in SRI001 by 5.2% and 11.9% respectively, and the fact that the high density areas are enhanced, is a strong indicator that these rise in links and tracks includes correct additions.
|
|---|
