AI System Helps Predict Outcomes for Patients Diagnosed with Epithelial Ovarian Cancer

AI System Helps Predict Outcomes for Patients Diagnosed with Epithelial Ovarian Cancer

A team of researchers from University College London and the University of Melbourne have developed AI software that can predict ovarian cancer prognosis and identify the likely effectiveness of treatments with four times the accuracy of current methods.

Epithelial ovarian cancer (EOC) has the highest mortality rate of all gynecological cancers, partly due to the difficulty in diagnosing the cancer in its early stages when treatment is most likely to be effective. Typically, EOC is only diagnosed once symptoms appear, which is not until the latter stages of the disease. EOC has a five-year survival rate of around 40%.

The most common subtype of EOC, and also the most-deadly, is high-grade serous ovarian cancer (HGSOC) which affects around 70% of patients diagnosed with EOC. Genetic profiling and microRNA data have been used to identify biomarkers to help diagnose the disease and determine EOC risk, although it is challenging to use those in a clinical setting due to intra-tumor heterogeneity, the time taken to perform analyses, and the high cost of assays.

Currently diagnosing EOC requires a blood assay for the biomarker CA125 and a CT scan to identify tumors and determine how far the cancer has progressed. Depending on whether the cancer has spread, surgery may be an option along with chemotherapy. Unfortunately, while CT scans can detect tumors, they cannot provide any information about how effective treatments are likely to be.

What is needed is a non-invasive, cost-effective, real-time, prognostic marker approach to help identify treatments that are likely to be most effective. That is where it is hoped the new AI system, termed TexLab, may be able to help.

TexLab uses machine learning to assess ovarian tumors based on their structure, shape, size, and genetic makeup, all of which have a bearing on survival rates. The analysis produces a Radiomic Prognostic Vector (RPV) score which provides an indication of the severity of the disease and the likely prognosis.

In a trial of the software at the UK’s Hammersmith Hospital, the researchers used CT scans from 364 EOC patients which were tested to 657 quantitative mathematical descriptors. The researchers also evaluated protein expression and genomic profiles from a sample of patients along with fresh frozen tissue samples to obtain histological, protein, and gene expression data to help assess the RPV results.

The researchers report that TexLab was four times as accurate at predicting death from EOC than current methods such as blood tests and prognostic scores. A high RPV score accurately predicted poor surgical outcomes and tumor resistance to chemotherapy.

“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,” said the authors.

The AI analysis is quick and inexpensive. All that is required is for CT scans to be fed into the system, which are immediately available. The system can calculate the RPVs of 80 EOC datasets in less than 5 minutes using a standard computer.

The researchers hope that the AI software can help inform treatment decisions based on how patients are likely to respond. The software could be used to help devise personalized treatment options for women diagnosed with the disease to improve survival rates.

The researchers are currently planning a much larger study to determine the accuracy of the software at predicting how patients will respond to different treatments and probable surgical outcomes.

Further information on the AI system and the results of the trial are detailed in the paper – A mathematical-descriptor of tumor-mesoscopic-structure from computed-tomography images annotates prognostic- and molecular-phenotypes of epithelial ovarian cancer – which was published in Nature Communications, 10, Article number: 764 (2019).

Leave a Reply