AI Model Combines Blood Test and CT Scan Analysis to Predict Therapy Responses in Ovarian Cancer Patients

Ovarian cancer annually impacts thousands of women, with many diagnoses occurring at advanced stages due to subtle early symptoms. High-grade serous ovarian carcinoma, which accounts for 70-80% of ovarian cancer cases, is particularly aggressive and often resistant to chemotherapy. Current methods for predicting response to therapy in these tumors are only about 50% accurate. The complexity and diversity of the disease among individuals have made it challenging to find reliable biomarkers. Now, researchers have developed an artificial intelligence (AI)-based tool to improve the accuracy of predicting chemotherapy responses in patients with ovarian cancer.

The tool, named IRON (Integrated Radiogenomics for Ovarian Neoadjuvant therapy), was developed by researchers at the Catholic University of the Sacred Heart (Milan, Italy). IRON analyzes a range of clinical features, including circulating tumor DNA from blood samples (liquid biopsy), patient demographics (age, health status, etc.), tumor markers, and CT scan images. It then predicts the likelihood of a successful therapy outcome, specifically the volumetric reduction of tumor lesions. Impressively, IRON can predict therapy outcomes with an 80% accuracy rate, a significant improvement over existing clinical methods.

For their research, the team compiled two datasets comprising 134 patients in total, with 92 in the first dataset and 42 in a separate validation set. They collected comprehensive clinical data for these patients, including demographic information, treatment specifics, blood biomarkers like CA-125, and circulating tumor DNA. Additionally, they gathered quantitative details from CT scans of all primary and metastatic tumor sites. Notably, omental and pelvic/ovarian sites, where ovarian cancer commonly spreads, were observed to carry the majority of the disease burden initially. It was found that omental deposits responded better to neoadjuvant therapy compared to pelvic disease.

The researchers also examined tumor mutations (such as TP53 MAF in circulating DNA) and the CA-125 marker in relation to the overall disease burden before treatment and response to therapy. Advanced analysis of CT scan images identified six patient subgroups, each with unique biological and clinical features indicative of their response to therapy. These tumor characteristics were fed into AI algorithms, creating a comprehensive model. After being trained, the model’s effectiveness was validated using the independent patient sample, showcasing its potential to enhance ovarian cancer treatment strategies.

"From a clinical perspective, the proposed framework addresses the unmet need to early identify patients unlikely to respond to neoadjuvant therapy and may be directed to immediate surgical intervention," said Professor Evis Sala who coordinated the study.

Related Links:
Catholic University of the Sacred Heart