Multiparametrik Prostat Manyetik Rezonans Görüntülemenin Uzun Dönem Takip Sonuçları ve Negatif Prediktif Değeri

Abstract

Multiparametric prostate MRI (mp-MRI) is an effective imaging modality that is frequently used in the early diagnosis of prostate tumors, pre-treatment staging and post-treatment follow-up. Our aims in this study are to examine the diagnostic performance of mp-MRI by using PI-RADS v2, to calculate the negative predictive value (NPV) in different clinical subgroups, to try to improve the diagnostic performance of mp-MRI with various models, to reveal the prognostic importance of PI-RADS v2, to evaluate “long-term follow-up results” of the patients with initial low PI-RADS scores (PI-RADS≤3) and to contribute to the patient management with our findings. In our study, clinical, biochemical, radiological and histopathological data of 2456 adult patients who underwent mp-MRI in Hacettepe University Department of Radiology due to the suspicion of prostate cancer (PCa) were collected retrospectively and a general data set was created. In the first part of the study, by using histopathologically-proven cases (810 patients and 826 biopsies), the diagnostic and prognostic performance of mp-MRI and PI-RADS v2 were evaluated, subgroup analyzes were performed and the performance metrics of machine learning methods were examined. In the second part, patients who have an initial PI-RADS score of ≤3 and adequate clinical follow-up (385 patients in the PI-RADS 1-2 group, 478 patients in the PI-RADS 3 group) were filtered from the general data set, and cancer-free follow-up probabilities were calculated by performing survival analysis. Considering cases with ISUP≥2 in histopathological examination as clinically significant prostate cancer (cs-PCa) and using PI-RADS v2, the NPV of mp-MRI was between 92.8% and 95.5%, while the positive predictive value (PPV) was about 33.2% to 44.8% in our study. Mp-MRI has high sensitivity (90.8%-98.4%) but low specificity (14.6%-51.6%). In PI-RADS 1-2, PI-RADS 3, PI-RADS 4 and PI-RADS 5 cases, cs-PCa detection rates were 4.6%, 8.1%, 36.4% and 60.6%, respectively. It is also found that the higher PI-RADS scores, the higher the probability of cancer with a poor prognosis. Unfavourable PCa ratios in PI-RADS 1-2, PI-RADS 3, PI-RADS 4 and PI-RADS 5 cases were found to be 6.8%, 9.9%, 31.5% and 65.6%, respectively. In patients with PI-RADS 3 lesions, "presence of previous negative biopsy" and "low PSA density (mPSAD<0.15 ng/mL2)" were found to be two main protective variables, both diagnostically and prognostically. In subgroup analyzes, both the probability of having cs-PCa and the likelihood of unfavourable PCa were less than 5% in these two PI-RADS 3 subgroups, showing similar results to the PI-RADS 1-2 group. Machine learning algorithms showed promising performance in predicting the probability of cs-PCa. The hybrid model created by the combination of several machine learning methods accurately classified approximately 90% of all cases with 88.2% NPV and 91.1% PPV. In the PI-RADS 1-2 group, 1-, 3-, and 5-year cs-PCa diagnosis-free survival probabilities were 99.1%, 96.5%, and 93.8%, respectively. As for the PI-RADS 3 group, 1-, 3-, and 5-year cs-PCa diagnosis-free survival probabilities were found to be 94.9%, 90.9%, and 89.1%, respectively. While mPSAD had a significant effect on cancer-free survival for both subgroups, the presence of previous negative biopsy had an impact on cancer-free survival only in the PI-RADS 3 group. In conclusion, mp-MRI has high NPV and sensitivity. On the other hand, its PPV and specificity are low. Yet, it is possible to improve PPV and specificity remarkably with the help of machine learning, while largely maintaining the already high NPV of mp-MRI. In “PI-RADS 1-2 cases”, “PI-RADS 3 cases with mPSAD<0.15 ng/mL2” and “PI-RADS 3 cases with previous negative biopsy”, clinical follow-up approach can be preferred over immediate biopsy since probabilities of cs-PCa and unfavourable PCa seem to be low. Moreover, it is useful to keep in mind that cs-Pca diagnosis-free survival will already be high in these groups, even during clinical follow-up. 3-year cancer-free survival probabilities are over 94% in all three groups.