Machine learning can differentiate types of pancreatic cancer tumors
An algorithm can identify from MRI images different subtypes of pancreatic cancer, aiding clinicians who plot prospective treatments.
Researchers say that knowledge of the specific type of a patient’s cancer may help pinpoint what kind of chemotherapy would be most effective to combat the disease.
Pancreatic ductal adenocarcinoma (PDAC) has the worst prognosis rate of all cancer tumors. Complete resection, often combined with chemotherapy, is still the only curative therapy option. However, a substantial number of patients doesn’t respond to chemotherapy or quickly become resistant to it. In addition, different types of chemotherapy work better on certain PDAC subtypes, so learning which subtype a patient has can greatly affect the patient’s outcome.
It is possible to differentiate the subtypes by use of molecular profiling. However, such profiling relies on tissue biopsies, which is expensive and time consuming and requires high tissue quality. Thus it is not routinely performed. However, noninvasive diffusion weighted MRI, which measures the random motion of water molecules, can quantify tissue microstructures well and is part of many routine diagnostic workups.
Researchers from the School of Medicine, Technical University of Munich, Germany, hypothesized that machine-learning based analysis of radiomic features of the tumors by reading MRIs could determine subtypes of PDAC.
“Pre-therapeutic identification of specific subtypes in pancreatic cancer is urgently required to guide individual treatment decision,” the researchers said.
The researchers developed an algorithm capable of predicting clinically relevant PDAC subtypes from pre-operative MRIs. They assessed 55 surgical PDAC patients, training the algorithm on 70 percent and testing on the remaining 30 percent.
The radiomic analysis of the images paired with the machine learning could discriminate with “high sensitivity and specificity” between two groups of molecular subtypes of PDAC, which have different responses to commonly used chemotherapy regimens. They also found that the algorithm ranked entropy as the most important radiomic feature evaluated.
The researchers additionally conducted follow up on the patients for six years to compare subtype, chemotherapy response and patient survival rates. The molecular subtype was “significantly” associated with overall survival.
The study was published October 2 in PLOS One.
“These findings could have tremendous implications in patient stratification and subtype-guided therapy selection. In addition, targeted therapies … are highly effective yet even more specific for a certain molecular profile and many new targeted, stroma- and immune-based treatment strategies are being explored. This increasing complexity requires robust and cost-efficient tools for clinically relevant patient stratification to best leverage current knowledge and advance the field,” the study authors stated.