Using Machine Learning, a team of scientists has developed a comprehensive test to help better treat patients for pancreatic cancer. The test, called CompCyst, significantly outperformed current gold standards-of-care on a test group of 875 patients.
Pancreatic cysts can be fatal, but not always
Pancreatic cysts are fluid-filled lesions in the pancreas. The pancreas is a vital organ, located behind the stomach, that produces hormones and enzymes to help in the digestion of food.
The formation of cysts tends to be found in up to 8% of all people over the age of 70.
Although most cysts are benign and tend to not cause any noticeable symptoms. But some pancreatic cysts that produce mucin can transform into an aggressive form of pancreatic cancer.
"They're typically found during imaging testing for another problem. Some are actually noncancerous (benign) pockets of fluids lined with a scar or inflammatory tissue, not the type of cells found in true cysts (pseudocysts)," according to the Mayo Clinic.
To date, it is difficult to distinguish precancerous cysts from benign ones. This means that noncancerous cysts are often misclassified and unnecessarily removed with pancreatic surgery.
This misclassification often leads to unnecessary invasive surgery to remove the cysts regardless of whether they are benign or more dangerous forms.
The new test could help avoid unnecessary surgery
Because of this the researchers, Simeon Springer, and his colleagues decided to see if they could improve triage processes for patients. They enrolled 875 patients with pancreatic cysts and collected information of the mutations, proteins, and other markers linked to their either benign or mucin-producing cysts.
The new test, using machine learning algorithms, some interesting findings were uncovered. From the 875 patient study, CompCyst revealed that 60% of them could have avoided surgery.
This is significant and would lead to fewer unneeded surgeries, effectively reducing medical costs and complications that are often associated with any form of invasive surgery.
CompCyst outperformed standard-of-care diagnosis
The team, using deep machine learning techniques, trained CompCysts to read markers from the data collected from enrolled patients. The system quickly classified patients into those that should be monitored, not monitored, or receive surgery.
CompCyst was trained using 436 of the original patients, and the researchers found the test largely outperformed standard-of-care pathology when evaluated in 426 other patients.
CompCyst correctly identified that 60.4% of patients should have been discharged. This was significantly great that 18.9% discharged using standard-of-care diagnosis.
48.6% of patients were also correctly classified as being in need of monitoring, versus 34.3% from traditional triage. Finally, and most importantly, it was able to identify 90.8% of patients in need of surgery versus 88.8% from regular industry gold standard diagnosis.
As promising as this is, future work will be necessary to prospectively validate the markers used in the test. Springer et al. say their platform has strong potential to be used in the clinic as a complement to existing approaches.
The original study was published in the journal Science Translational Medicine.