A novel artificial intelligence (AI) approach, combining cell-free DNA (cfDNA) fragment patterns with levels of the proteins CA125 and HE4, shows improved accuracy in detecting ovarian cancer. The AI-enhanced model, leveraging machine learning, can differentiate patients with ovarian cancer from healthy individuals or those with benign ovarian masses, potentially addressing the challenges of late diagnosis and improving survival rates.
Enhancing Ovarian Cancer Diagnosis with AI
Ovarian cancer, the fifth most common cause of cancer deaths among women in the United States, often presents without symptoms in its early stages, leading to late diagnoses and a five-year survival rate of approximately 50%. Researchers have been exploring liquid biopsy technologies, analyzing blood for tumor-derived DNA, to noninvasively detect various cancers. A new method of liquid biopsy analysis, called fragmentomics, has shown promise in improving the accuracy of such tests.
Jamie Medina, PhD, a postdoctoral fellow at the Johns Hopkins Kimmel Cancer Center, explained that DELFI (DNA Evaluation of Fragments for early Interception) utilizes fragmentomics to detect changes in the size and distribution of cfDNA fragments across the genome. "Because cancer cells are rapidly growing and dying and have chaotic genomes as compared to healthy cells, patients with cancer have different patterns of DNA fragments in their blood than patients without cancer," Medina stated.
AI Model Performance and Validation
Medina, Akshaya Annapragada, and colleagues trained 20 AI predictive models using 6,778,762 laboratory results screening for ovarian cancer. Each predictive model was trained to produce an accurate diagnosis of ovarian cancer, and the model’s decision-making was based on 52 biomarkers. The predictions made by the model were validated against three datasets of laboratory results. The performance of the model in accurately predicting positive cases exceeded laboratory tests with and without the conventional carbohydrate antigen 125 (CA125) and human epididymal protein 4 (HE4) biomarkers most commonly associated with ovarian cancer. Importantly, these capabilities included the improved prediction of early-stage ovarian cancer, a result which frequently eludes traditional testing methods due to deficiencies in the specificity and sensitivity of current biomarker tests.
The researchers analyzed plasma from 134 women with ovarian cancer, 204 women without cancer, and 203 women with benign adnexal masses. At a specificity of over 99%, the screening model identified 69%, 76%, 85%, and 100% of ovarian cancer cases staged I-IV, respectively, with an area under the curve (AUC) of 0.97 across all stages. In comparison, an analysis of CA125 levels alone identified 40%, 66%, 62%, and 100% of cases staged I-IV, respectively. The diagnostic model differentiated ovarian cancer from benign masses with an AUC of 0.87.
Clinical Implications and Future Directions
Victor Velculescu, MD, PhD, FAACR, senior author of the study, emphasized the significance of these findings: "Our goal was to overcome this challenge by combining genome-wide cell-free DNA fragmentation with protein biomarkers to develop a new high-performance approach for early detection of ovarian cancer." The group intends to validate their models in larger cohorts to strengthen the associations observed.
The predictive model developed by Cai and colleagues serves as a promising first step toward strengthened capabilities in the accurate, timely diagnosis of ovarian cancer. In order to ensure access to the model for public review and potential utilization, Cai and colleagues have published it as an open-source tool on the internet. As expected for other clinical applications of AI, the increased adoption of AI-enhanced ovarian cancer detection could have resounding effects on patients and clinicians, but also on the ovarian cancer epidemiological landscape more broadly. While the incident cases of ovarian cancer could increase due to higher diagnosis rates, strengthened early detection could shift the stage distribution of diagnosed cases toward earlier stages, and potentially lead to improved odds of survival.
Limitations of this study include a relatively small sample size, a study population primarily comprised of American and European patients, and the retrospective nature of the analysis.
