AI Model Trained on Pre-Treatment Laparoscopic Surgical Videos Predicts Treatment Outcomes in Ovarian Cancer

The study by researchers at the University of Texas MD Anderson Cancer Center (Houston, TX, USA) trained an AI model to use still-frame images from pre-treatment laparoscopic surgical videos to predict outcomes in two predefined populations of patients with high-grade serous ovarian cancer (HGSOC): those with excellent response (ER) to standard treatment and those with poor response (PR) to standard therapy.

The study examined videos from 113 HGSOC patients, 75 (66%) of whom had a durable response to the therapy (ER). A total of 435 still-frame images from four anatomical locations – diaphragm, omentum, peritoneum and pelvis – were used to develop the AI model to detect distinct morphological patterns of disease in the patients, correlate those patterns with outcomes, and discriminate between the two patient populations (ER or PR). The images were divided into three sets: 70% for training, 10% for validation and 20% for testing. The model effectively predicted outcomes with an overall accuracy 93%. It successfully identified all patients with ER but misclassified about one-third of patients with PR as ER patients, possibly because of the smaller number images available for these patients in the study.

“This pilot study is an exciting frontier in surgical innovation that shows how we can use machine learning to enhance our clinical approach to treating patients with gynecologic cancers,” said Deanna Glassman, MD, The University of Texas MD Anderson Cancer Center, who co-led the study. “A major implication of our study is that the AI model could identify patients who are likely to have a poor response to traditional therapies, enabling clinicians to alter surgical plans and goals, and providing opportunities for tailoring therapeutic strategies in those patients.”

“The concept of using an AI model trained on laparoscopic images requires additional validation studies, but in the future it could be extended to other gynecologic cancers to identify patterns of disease, predict treatment outcomes, and distinguish between viable and necrosed malignant tissue at the time of interval debulking surgery (IDS),” Glassman said.

Related Links:
University of Texas MD Anderson Cancer Center