The MetNet.

Top Line: Peritoneal metastases are never good and usually signal one to make a turn towards systemic versus aggressive local-only therapy.

The Study: What’s more, as many as two-thirds of advanced gastric cancers have them. Unfortunately, given the insidious nature of their presentation, peritoneal mets aren’t typically easy to detect until the patient is open on the table. In 2021, where should we turn for a solution? A convolutional neural network, of course, in this case the Peritoneal Metastases Network (aka the PMetNet) developed by a savvy group of Chinese physicians and tech whizzes. It retrospectively trained on pre-operative CT scans obtained on 1225 patients undergoing resection for gastric cancer at a single tertiary care center in Guangzhou, China. It was then prospectively validated on 753 cases treated at two other tertiary care centers also in Guangzhou. Since all patients went to the operating room, the gold standard here was pathologic confirmation of peritoneal mets. The training model predicted peritoneal mets on pre-op CT with a sensitivity of 75% and a specificity of 93%. The validation model did even better with an impressive sensitivity of 88% and a specificity of 98%. Wow. Given these numbers, it’s no surprise that, on a multivariable analysis of clinical features, the PMetNet was a strong independent predictor of peritoneal mets.

TBL: The PMetNet may represent a huge breakthrough in avoiding the common scenario of patients with advanced gastric cancer and peritoneal mets undergoing a futile invasive procedure. | Jiang, JAMA Netw Open 2021