Radiomics and Machine Learning in the Diagnosis of Ovarian Masses

The correct differential diagnosis between benign and malignant adnexal masses is the main goal of preoperative ultrasound diagnostics and is very important to plan the correct treatment for the patient in terms of surgical team (gynecologist oncologist or benign pathology center), surgical access (laparoscopy / laparotomy) and type of surgery (conservative / demolitive). Several ultrasound models have been developed to help gynecologists define the risk of malignancy of adnexal masses. In order to use the predictive models, the examiner had to collect certain ultrasound features of the lesion which, integrated with the patient's clinical and / or biochemical characteristics, provided a risk of malignancy of the mass. Recently radiomics is emerging as an interesting tool to be applied to diagnostic imaging (computed tomography, magnetic resonance and even ultrasound). Radiomics is the evaluation of images through complex software that allows to 'read' the intrinsic characteristics of the tissue identifying aspects that are not visible by subjective interpretation of the operator, in a fully automated and therefore reproducible way. Radiomics applied to artificial intelligence for the creation of predictive models represents an interesting tool to overcome the limitations of previous models, at least partly dependent on the operator's experience. Among the serous ovarian cancer, those with BRCA gene mutation represent an interesting subgroup and are characterized by a different pathophysiological history than wild type tumors due to greater chemosensitivity and the possibility of targeted treatment with antiangiogenic drugs and PARP-inhibitors. The application of radiomics to preoperative ultrasound images could identify BRCA mutated tumors before surgical planning (radiogenomic analysis) and allow a personalized treatment. The aim of the study is to validate a predictive model to define the risk of malignancy of adnexal masses that the investigators developed at the Fondazione IRCCS Istituto Nazionale dei Tumori di Milano. The model, based on the integration of radiomics and artificial intelligence, uses complex software capable of 'reading' the ultrasound images in a completely automatic way and is able to estimate the risk of malignancy of the mass. If the patient decide to participate in the clinical study, the patient will undergo transvaginal ultrasound (eventually supplemented by transabdominal ultrasound in case of large adnexal masses, if the patients are virgo or if the patients will refuse transvaginal approach for any reason). This exam is part of the routine preoperative evaluation for adnexal pathology and therefore the patients don't have to undergo any additional clinical, biochemical or imaging examination, according to national and international guidelines. Thereafter, the images stored during the preoperative ultrasound will be exported in anonymous format from the ultrasound system, and sent to the coordinating center (Fondazione IRCCS Istituto Nazionale dei Tumori di Milano). There, images will be submet to radiomic analysis through the application of a dedicated software; that will allow to evaluate the intrinsic characteristics of the tissue according to different parameters (shape, intensity, grade of heterogeneity and many others) of the 'pixels' (gray dots) that constitute the ultrasound image. This analysis, once validated, will provide clinicians an additional tool to identify malignant adnexal masses prior to surgery. If the final histological diagnosis is of serous epithelial ovarian cancer, through the use of the same radiomics software described above the investigators will try to identify the intrinsic characteristics of the tissue associated with the presence or absence of the BRCA 1 or 2 mutation

Sex

Female

Minimum age

18 Years

Inclusion Criteria:

* Patients scheduled for surgery for adnexal masses
* Transvaginal ultrasound available within 6 weeks before surgery
* Age \>17 yeras

Exclusion Criteria:

* Pregnancy
* Consent withdraw