Tese de Doutoramento Medicina 2024 Faculdade de Ciências Médicas, Universidade NOVA de Lisboa

Prostate cancer (PCa) is one of the most common cancers in men, showing an increasing incidence in most of Europe and the United States, mainly due to improved diagnostic capabilities. In the last 10 years, the approach to prostate cancer has undergone significant changes with the implementation of multiparametric MRI (mp-MRI) as a diagnostic and prognostic tool, as well as for detecting tumour aggressiveness, which impacts clinical decision-making for patients. However, regarding the staging of this tumour, MRI is still the subject of scientific research and is not universally accepted as the reference test for local staging of PCa in all patients. Nevertheless, in our clinical practice, we firmly believe that MRI plays a crucial role in differentiating between localized (confined to the prostate) and locally advanced tumours with extra-prostatic extension. This aspect is essential in deciding the appropriate therapeutic strategy to apply. But how can MRI help differentiate patients with PCa who will have extracapsular extension (ECE)? What criteria on the MRI can identify ECE, and are they capable to detect microscopic ECE? Are they reproducible among different observers? Furthermore, can MRI, when combined with artificial intelligence techniques, provide additional information about the tumour biology that is not apparent in visual analysis and can reduce variability between observers? Do interpretative variables have post surgical prognostic value even in low-risk patients? These were the questions that led to this scientific project, which consisted of a longitudinal observational study in patients with PCa, evaluated by MRI before radical prostatectomy, and it was divided into three phases: 1 - Construction of the semantic model (MRI interpretation): This stage was based on visual diagnosis of MRI by the radiologist combined with clinical and histological variables, aiming to build a statistical interpretive model capable of predicting ECE. The researchers obtained a semantic statistical model based on four predictive variables: the Gleason score from prostate biopsy and three findings on MRI: tumour capsular contact, capsular disruption, and visualization of extra-prostatic tissue on MRI. This model showed good accuracy and sensitivity, with slightly lower specificity and acceptable reproducibility. The researchers identified biomarkers like capsular contact and more aggressive Gleason scores, which were more suitable for evaluating ECE at earlier stages. 2 - Construction of the radiomics model: Another predictive model was built based on statistics extracted from the images (radiomics) to assess ECE and compared it with other models using clinical variables and MRI interpretation (semantic), using both classic and more recent metrics adopted in Machine Learning, aiming to innovate in interpretation and potential clinical applications of this algorithm. The researchers concluded that, in their sample, the radiomics model did not show a higher AUC than the semantic model, and the combined model (radiomics plus semantic) would be the most suitable for evaluating the surgical decision in terms of preserving periprostatic nerve plexus. However, there were no significant discrepancies between the metrics of the two models. 3 - Evaluation of the impact of clinical and MRI variables on post-surgical biochemical recurrence-free survival. In this phase, the researchers assessed how the clinical and MRI variables used in the first phase, along with histological aspects obtained from the prostatectomy specimen, determined biochemical recurrence-free survival within a maximum of 4 years after surgery. In addition to the important role of pathologic tumour stage as a prognostic factor, the presence of macroscopic ECE, high tumour contact length (TCCL), capsular disruption used for detecting extracapsular extension before surgery, also have a significant impact on biochemical recurrence and should be taken into consideration in clinical decision-making. Through this project, the researchers were able to conclude the following: MRI is an important imaging technique, should be done for staging all patients before surgery, as it can accurately diagnose ECE, even in early stages, when combined with the Gleason score of the tumour, increasing the surgeon's confidence in deciding the type of surgery to perform. The radiomics model combined with semantic evaluation may have some role in the surgical decision, but it did not prove to be individually superior to the interpretive model. The predictive features of MRI for detecting extracapsular extension before surgery are also considered independent prognostic factors for early biochemical recurrence enabling the identification of low/intermediate-risk patients who may require more intensive follow-up and potentially early intervention strategies