A score derived from CT imaging using a machine learning radiomics approach was able to reliably identify epithelial ovarian cancer patients with poor survival outcomes, according to a new study. The method could be used to guide a more personalized approach to therapy in this malignancy, and is potentially transferable to other cancers as well.
Radiomics involves quantifying tumor characteristics based on imaging and advanced bioinformatics tools. It could be useful in epithelial ovarian cancer, given the heterogeneity of response to therapy; also, recent research has used molecular subtyping to try and determine risk profiles for epithelial ovarian cancer. “It remains challenging, however, to translate these molecularly determined characteristics into clinically relevant biomarkers due to intratumor heterogeneity, additional high assay cost, and time delays,” wrote study authors led by Haonan Lu, of Imperial College London. “Therefore, a noninvasive, real-time, and cost-effective prognostic marker approach is warranted to reliably guide personalized treatment of epithelial ovarian cancer patients.”
The researchers developed a program that summarized 657 features from contrast-enhanced CT scans of 364 epithelial ovarian cancer patients at their initial presentation; these features related to the shape, size, intensity, texture, and wavelet decompositions of the scans. Using a machine learning approach, they derived a statistic they termed the Radiomic Prognostic Vector (RPV). The study was published in Nature Communications.
In a discovery dataset, RPV was continuously associated with overall survival, with a hazard ratio of 3.83 (95% CI, 2.27–6.46; P = 5.11 × 10-7). This was also true in two separate validation datasets, and RPV fared better than some existing prognostic markers—including C125 and a transcriptome-based molecular subtype—in predicting survival. Along with survival, a high RPV was also associated with primary chemotherapy resistance, shorter progression-free survival, and poor surgical outcomes.
“We demonstrate, based on the strong association between RPV and response to primary chemotherapy or surgery, that patients with high RPV have a significantly high risk of failing quality surgery or systemic strategies and suggest that they possibly need to be directed towards alternative therapeutic approaches including stroma modifying therapies,” the authors wrote. The tool, they added, “convincingly fulfills an unmet need” in the setting of epithelial ovarian cancer.
The field of machine learning–guided radiomics is rapidly expanding, in this and other malignancies. “Radiomics appears to offer a nearly limitless supply of imaging biomarkers that could potentially aid cancer detection, diagnosis, assessment of prognosis, prediction of response to treatment, and monitoring of disease status,” wrote several experts led by Robert J. Gillies, PhD, of the department of cancer imaging at the H. Lee Moffitt Cancer Center in Tampa, Florida, in 2015.
And since then, that potential has started to be realized further: at present, 38 trials involving radiomics are active or recruiting, according to ClinicalTrials.gov. These include trials in breast, lung, colorectal, esophageal, and many other malignancies. In one study using radiomics in lung cancer published in late 2018, authors led by Ahmed Hosny, of Dana-Farber Cancer Institute, wrote, “Deep learning algorithms that learn from experience offer access to unprecedented states of intelligence that, in some cases, match human intelligence. … This emerging approach allows for early diagnosis and customized patient-specific treatments, thus delivering the appropriate medical care to the right patient at the right time.”