Advanced ovarian cancer surgical planning often begins long before a patient enters the operating room. The decision to proceed directly to surgery or to start with chemotherapy depends largely on what physicians see on preoperative imaging. In ovarian cancer, that picture is not always complete.

Dr. Behrouz Zand, a gynecologic oncologist at Houston Methodist, is leading a collaborative project with Rice University to change that reality through artificial intelligence. Supported by a seed fund award from the Houston Methodist-Rice University collaboration, the team is developing a machine learning model to provide surgeons with a more accurate assessment of disease burden and surgical complexity before an operation begins.

Looking beyond the CT scan

For patients with advanced-stage ovarian cancer, complete cytoreduction remains one of the strongest predictors of improved outcomes.

"Removing all visible cancer is strongly associated with better outcomes," says Dr. Zand. "The problem is that the disease can be difficult to fully characterize on CT scans."

Tumor deposits may be hidden in anatomically challenging locations or appear less conspicuous on imaging studies. As a result, surgeons occasionally encounter unexpected findings in the operating room, including disease that cannot be safely or completely resected.

"Every oncologist who treats ovarian cancer has experienced a situation where a patient is taken to surgery and unexpected findings are discovered," Dr. Zand says. "Not all the tumor can be removed, and those patients tend to have worse outcomes."

The goal of the project is not to replace physician judgment but to provide clinicians with additional data to support preoperative decision-making.

Combining clinical data and imaging

The research team is developing a multimodal machine learning pipeline that integrates clinical information, laboratory values, biomarkers and CT imaging data.

Houston Methodist investigators provide clinical and surgical expertise, while collaborator Dr. Meng Li of Rice University's Department of Statistics leads the advanced statistical modeling and AI development.

Together, the team has already created a clinical database and completed preliminary modeling. The next phase focuses on analyzing CT DICOM images using radiomics, which extracts quantitative features from imaging data that may not be readily apparent to clinicians.

[blockquote]

The team believes that integrating multiple data sources may enable the model to identify patterns associated with tumor location, resectability and surgical complexity.

Building a surgical roadmap

Dr. Zand compares CT imaging to a map. Artificial intelligence, he says, may add another layer of information to help surgical teams navigate the journey ahead more effectively.

The researchers hope the model will ultimately predict factors such as:

• Likelihood of complete tumor resection

• Surgical complexity

• Estimated operative time

• Anticipated blood loss

• Length of hospitalization

"If we can predict what a surgeon may encounter in the operating room, it could help with planning and preparation," Dr. Zand says.

The work focuses on patients with stage III and stage IV ovarian cancer, where treatment decisions are often complex and the stakes are high.

Training the model

The current study is retrospective, drawing on data from around 90 ovarian cancer patients treated at Houston Methodist. These cases are used to train and refine the algorithm.

At this stage, Dr. Zand emphasizes that the dataset is not yet large enough to support treatment decisions. Instead, the team is focused on establishing feasibility and assessing how accurately the system can predict operative findings.

"So far, what we've learned is that this pipeline is feasible," he adds. "But it really depends on how well you train it. The more patients you have, the better."

Once the model is more fully developed, the investigators plan to pursue validation across multiple institutions to expand the dataset and improve its performance.

A tool for surgeons, not a replacement

As AI continues to gain traction across medicine, Dr. Zand is careful to distinguish this effort from publicly available generative AI platforms.

The model under development is an in-house machine learning system designed to address a clinical problem in ovarian cancer surgery. It's not intended to make decisions independently but to provide physicians and patients with more information.

"The overall goal is not to let the algorithm make the decision," Dr. Zand says. "It's to give patients and their surgeons better information before a major cancer operation."

If successful, the technology could improve patient counseling, refine surgical planning and support a more personalized approach to care.

"Better planning, better counseling and more personalization for the patient ultimately lead to better outcomes," Dr. Zand says.

While the current project focuses on advanced ovarian cancer, the concept may have broader implications. A predictive platform that integrates imaging and clinical data to forecast surgical complexity could eventually be adapted to other cancer types and procedures.

For now, however, the team's focus remains on providing gynecologic oncologists with a more complete picture of disease before they ever pick up a scalpel.