Prediction of Ovarian Cancer Histotypes and Surgical Outcome

The standard treatment for advanced ovarian cancer (AOC) is primary cytoreductive surgery (PCS) followed by adjuvant chemotherapy. However, optimal cytoreduction is not always achievable, particularly in cases with high tumor burden or patient frailty. In such scenarios, neoadjuvant chemotherapy (NACT) followed by interval cytoreductive surgery (ICS) represents a valid alternative with comparable oncologic outcomes in selected patients.

To optimize surgical strategy, objective tools are needed to identify the best candidates for PCS. Scoring systems such as the Fagotti Score and the Predictive Index Value (PIV) assess tumor resectability, but their accuracy largely depends on surgeon expertise.

It has already developed the PREDAtOOR project, a significant advancement in the use of artificial intelligence (AI) for managing AOC. PREDAtOOR has demonstrated high accuracy in both predicting the Fagotti Score and segmenting lesions from diagnostic laparoscopy videos, thus supporting a more objective and reproducible surgical decision-making process.

Importantly, therapeutic strategies should also consider tumor biology, as the response to NACT varies across histological and molecular subtypes. Unfortunately, such information is usually derived from histopathological and genomic analyses performed only after the surgical decision.

Kurman and Shih proposed a dualistic model of epithelial ovarian tumors, with distinct clinical and molecular features:

Type I tumors (low-grade serous, endometrioid, clear cell, mucinous): indolent growth, typically confined to the ovary, with stable genomes. Early-stage cases may be cured surgically. Metastatic Type I tumors tend to be chemoresistant but may respond to targeted therapies.

Type II tumors (high-grade serous carcinoma \[HGSC\], carcinosarcomas, undifferentiated carcinomas): aggressive behavior, marked genomic instability, and frequent homologous recombination deficiency (HRD). Although initially sensitive to platinum-based chemotherapy and PARP inhibitors, resistance often emerges.

Among these, HGSC is the most frequent and lethal. Yet, even within HGSC, substantial variability in chemotherapy response and clinical outcome is observed. A recent morphologic classification of HGSC stratifies tumors into infiltrative vs. expansive patterns, associated with specific molecular alterations and therapeutic responses.

However, these morphological and molecular features are not yet integrated into intraoperative decision-making, highlighting a need for new intraoperative tools to personalize care.

In this precision medicine landscape, AI, particularly through machine learning and computer vision, offers powerful solutions. These technologies can process large, heterogeneous datasets and automate intraoperative assessments, enhancing objectivity and diagnostic reproducibility. While AI-based classification of histologic and molecular subtypes from laparoscopy remains largely unexplored, it holds the potential to revolutionize treatment stratification in AOC.