No abstract

No abstract

we built this CAD system and we validated its reliability, we decided to test 10 patients with a biopsy confirmed PCa. All of them underwent MRI performing a conventional study with T1-w, T2-w and diffusion sequences. After precontrast acquisitions, patients were intravenously given gadobutrol and a total of 28 phases were acquired, each lasting 13 seconds. All images were examined by an expert radiologist on the basis of histological information provided after surgery and a total of 13 tumours, all located in the peripheral zone (PZ), were detected. A ROI was drawn around each lesion, on all possible slices to cover the whole tumour extension. When possible a ROI on a healthy region, with dimension comparable to that of the corresponding malignant ROI was also drawn for each patient. Information of pixels belonging to the same ROI were extracted from both T2-w sequence and the 28DCE volume acquisitions, to construct time-intensity curves over time. A filtering operation was performed to reduce noise contribution and signal to noise ratio was estimated to discard low quality data. T2-w images were used to evaluate mean grey value of pixels on selected ROIs, while DCE-MRI points were analyzed applying three different quantitative models (Tofts, Weibull, EU1) and a semiquantitative description (peak location and maximum enhancement, initial slope, curve wash-out, area under the curve). A total of 13 features were collected for each pixel. The initial features set was reduced in order to avoid over-fitting problems and to discard redundant information. Furthermore when a couple of highly correlated features occurred, the parameter of the couple with lower performances rate was discarded. On the basis of these elaboration steps a 6-dimensional vector was generated for all the pixels in which model fitting was successful. Malignancy probabilities were then calculated with the Bayes rule. Results: The resulting area under the receiver operating characteristic (ROC) curve was 0.874; sensitivity and specificity were 84.6% and 83.4% respectively. Good separation between malignant and benign points can be observed for the three combination of parameters shown on the Scatter plots of the three quantitative models implemented. Conclusion: The CAD scheme presented in this study shows good performance in discriminating between benign and malignant regions in the prostate. This system achieves a high sensitivity and specificity, leading to a better lesion detection rate. Future developments will focus on integrating the dataset with information from diffusion, in order to further improve system performances. RCC, N0, M0 between 1995 and 2007. TE was done in 332 patients, while RN in 143. Local recurrence, progression-free survival (PFS) and cancer-specific survival (CSS) were the main outcomes of this study. The KaplanMeier method was used to calculate survival functions, and differences were assessed with log-rank statistics. Univariate and multivariate Cox regression models were also used. Results: The surgical margin statu...

No abstract