AI enhances method for rapid diagnosis, potentially better treatment of endometrial cancer

Image

Artificial intelligence has helped researchers uncover potential avenues for better treatment of endometrial cancer, the most common gynecologic malignancy in the developed world and one with a worsening mortality rate, particularly among Blacks and Hispanics. 

With the aid of deep learning, a type of AI, a large multi-institutional collaboration including the Frederick National Laboratory for Cancer Research demonstrated a method of predicting genetic subtypes and mutations from histology, the physical characteristics of tumor tissue as seen under a microscope.  

Histology, which is relatively fast and cheap to perform, is commonly used to prioritize  cancer treatments. This is because the high cost of genomic techniques combined with a scarcity of insurance coverage for these analyses in early-stage disease limits their use at a time when doctors are deciding an initial course of treatment. 

“We therefore explored the possibility that features derived from traditional histopathology images could be used to predict the molecular features of a tumor,” members of the National Cancer Institute’s Clinical Proteomic Tumor Analysis Consortium noted in the September 11 issue of Cancer Cell.  

The scientists compared 138 endometrial cancer tumor samples with 20 normal tissue samples using 10 methods of analysis, including whole-genome sequencing, whole-exome sequencing, methylation array, total RNA sequencing, microRNA sequencing, targeted proteomics, global proteomics, phosphoproteomics, acetylproteomics, and glycoproteomics. They identified 1,292 upregulated genes and 1,488 downregulated genes within the collection. 

Using this latest information and building on earlier studies, the research group developed a new assay that could help determine which endometrial cancer patients would benefit most from going on immunotherapy, in which the immune system is primed to recognize and attack cancer cells.  

A second finding from the study is the possibility of measuring the activity of MYC (a cancer-causing gene) to determine if an endometrial cancer patient would benefit from metformin treatment. The drug has been used to treat type 2 diabetes but also is associated with lowering MYC activity in certain endometrial cancer patients, the researchers noted. 

The study also used machine learning to predict the molecular features of a tumor from images of biopsy tissue as seen under a microscope.  

“The high degree of accuracy achieved with our algorithm as a predictive model for some EC subtypes and mutations…is promising but needs to be tested on a larger cohort,” the scientists wrote. Collaborating on the research was Mathangi Thiagarajan of the FNL’s Clinical Research Directorate whose team collected the biospecimens used in the study.