Predicting Outcome of Cytoreduction in Advanced Ovarian Cancer, Using a Machine Learning Algorithm and Patterns of Disease Distribution at Laparoscopy (PREDAtOOR) (PREDAtOOR)

Predicting Outcome of Cytoreduction in Advanced Ovarian Cancer, Using a Machine Learning Algorithm and Patterns of Disease Distribution at Laparoscopy (PREDAtOOR)

Predicting Outcome of Cytoreduction in Advanced Ovarian Cancer, Using a Machine Learning Algorithm and Patterns of Disease Distribution at Laparoscopy (PREDAtOOR)

PREDAtOOR is a pilot study and this study aims at improving the selection of the best treatment strategy for patients with advanced ovarian cancer by using Camera Vision (CV) to predict outcomes of cyto reduction at the time of Diagnostic laparoscopy.

For the treatment of advanced ovarian cancer, the decision to undergo primary surgery is complex and decided by the surgeon while multiple considering multiple elements. Sometimes, chemotherapy is needed before surgery to shrink some of the tumours. To choose the best patients for primary surgery, several prediction tools have been developed. CT and MRI have most commonly been used to identify sites and amounts of tumors in the abdomen and can help determine if these tumours can be safely removed by surgery. However, these imaging methods are only a prediction, and sometimes a diagnostic laparoscopy (putting a camera in the abdomen to look at all sites of disease) is performed to help this decision process. With the introduction of artificial intelligence and machine learning, there is a possibility to create more precise prediction models using images from these diagnostic laparoscopy videos. In particular, the investigators would like to use images from the diagnostic laparoscopy to create machine-learning models to help predict if the tumours can be successfully taken out at primary surgery, or if chemotherapy before surgery would be needed. The investigators will enroll patients at a one-time point (being the time of surgery) and follow them forward in time and There will be no additional visits other than the surgery. During surgery time the surgical team takes images however, what makes this different is that these images will be used to help create an algorithm to predict surgical outcomes. These images will be stored in a secure database with an anonymous number not linking these pictures to any of the participants.

This study focuses on individuals diagnosed or suspected to have Stage III-IV ovarian cancer They must be fit for cytoreductive surgery These individuals also be selected for interval cytoreductive surgery after NACT

individuals who have been diagnosed or are suspected to have Clinical Stage III-IV Ovarian Cancer and CT and MRI have most commonly been used to identify sites and amounts of tumors in the abdomen and can help determine if these tumors can be safely removed by surgery. However, these imaging methods are only a prediction, and sometimes a diagnostic laparoscopy (putting a camera in the abdomen to look at all sites of disease) is performed to help this decision process.

With the introduction of artificial intelligence and machine learning, there is a possibility to create more precise prediction models using images from these diagnostic laparoscopy videos. In particular, it would like to use images from the diagnostic laparoscopy to create machine-learning models to help predict if the tumors can be successfully taken out at primary surgery, or if chemotherapy before surgery would be needed. During surgery time the surgical team takes images however, what makes this different is that these images will be used to help create an algorithm to predict surgical outcomes. These images will be stored in a secure database with an anonymous number not linking these pictures to any of the participants.

The Fagotti score, also known as the Predictive Index Value (PIV), is determined through the evaluation of six abdominal areas during laparoscopic exploration. These areas include the parietal peritoneum, diaphragm, greater omentum, bowel, stomach/spleen/lesser omentum, and liver. A score of 2 is assigned to each area with visible tumor spread, allowing for a maximum score of 14. Notably, a PIV score of 10 or higher signifies a threshold for triaging patients toward neoadjuvant chemotherapy. To create a predictive model for cytoreduction outcomes during diagnostic laparoscopy, advanced deep neural networks will be trained. This aims to automate PIV score assessment using a fully supervised approach and deduce features from images obtained during diagnostic laparoscopy to predict the possibility of a resection target above 1 cm or a lack of indication for cytoreductive surgery in a weekly supervised manner.

b)Number of Participants with Treatment Diagnostic Laparoscopy assessed by utilizing machine learning and computer vision models to analyze images and videos

The laparoscopic evaluation also demonstrated its efficacy in foreseeing surgical outcomes for patients undergoing interval cytoreductive surgery post neoadjuvant chemotherapy (NACT). However, this model remains vulnerable to the subjectivity inherent in each surgeon's evaluation of individual disease sites. Evaluating patients during intraoperative procedures during diagnostic laparoscopy often relies on a surgeon's judgment, which may not always be optimally trained for such evaluations and can be influenced by biases. Utilizing CV models can involve training them to automatically replicate expert assessments, providing more accurate evaluations for a larger patient population.

1. Number of Participants with treatment Diagnostic Laparoscopy assessed the images and videos by validating and/or updating an ML model.

The most promising machine learning (ML) models for preoperatively predicting cytoreduction outcomes have been recently identified through a systematic review. These models will undergo validation using the dataset and annotations gathered in this project. If required, the model will be further refined and updated to enhance its performance. Given that there are multiple variables of various natures (such as clinical characteristics, laboratory values, radiological features, and intraoperative findings) that impact cytoreductive surgery outcomes, ML models are well-suited for handling extensive sets of variables, particularly when the relationships between them are non-linear. The goal is to develop a predictive model for cytoreduction outcomes based on clinical characteristics, laboratory values, and radiological features.

Inclusion Criteria: Patients treated at Fondazione Policlinico Gemelli Hospital, Rome Italy, Trillium -Credit Valley Hospital, Mississauga, Ontario and Princess Margaret Cancer Centre, Toronto, Canada Patients fit for cytoreductive surgery Patients with a primary diagnosis of suspect Stage III-IV ovarian cancer Patients selected for interval cytoreductive surgery after NACT Exclusion Criteria: Patients with pre-operative Stage I-II disease confined to the pelvis Patients unfit for surgery Lack of information about patients' surgical outcomes and clinicopathological characteristics LGSOC, Clear cell and mucinous, non-epithelial histologic subtypes (if available)