Computer-aided diagnosis (CAD) na avaliação de nódulos prostáticos através da ressonância magnética multiparamétrica

Abstract

Context: Magnetic Resonance Imaging (MRI) of the Prostate is a recognized method for prostate cancer diagnosis, particularly when using a multiparametric protocol. However, to evaluate the large number of complex images requires trained radiologists and elevated amount of time for analysis, difficulting its application. In this regard, computational aid techniques for medical diagnosis emerged not only as a work support strategy, but also qualifying image characteristics that are not appraisable by the human eye. Eventually, such characteristics details may be decisive in clinical ascertainment. Objective: to identify the computer aided diagnostic (CAD) applied in Multiparametric MRI published knowledge for detection of prostate cancer, and, to compare it with the published meta-analysis data of PIRADS for its ascertainement. Data source: we used established expressions defined in MEDLINE/PUBMED, ScienceDirect, IEEE, Cochrane Library and CAPES Journal Portal to search bibliographic databases in English and Portuguese languages. Steps for studies selection: an evidence matrix was built with the relevant publications regarding the inclusion and exclusion criteria, as well as the quality filter. Duplicated publications or opinion articles were excluded by titles and abstracts review. Eligible studies were retrieved for full text lecture and evaluation. The evidence matrix titles were then indexed with the important study variables, and specific data was retrieved by interviewing each full text selected study. Data synthesis and results: 36 articles were selected in the CAD group, categorized according to the type of classifier (SVM, random forest, neural networks, amid others), and 10 studies in the PIRADS group. The available statistical variables were "area under the receiver operating" (ROC) curve, sensitivity, specificity. The respective CAD group means were 0.849; 0.799 and 0.796. In the PIRADS group, similar means were observed, with 0.838; 0.817 and 0.71 respective estimates. Mann-Whitney test of independent samples was applied, with no statistically significant difference being observed between groups. Conclusion: the detection of prostate cancer through CAD applied in Multiparametric Magnetic Resonance is of value, with no discrepancy when compared with the established scoring process (PIRADS), adding possible gains in terms of analyses time lenght and radiologists' effort.