OR WAIT null SECS
AI could help distinguish between head and neck and lung cancers, researchers found.
Under the microscope, both squamous cell carcinoma of the lung and squamous cell carcinoma of the head and neck appear as dense cell groups with non-specific growth patterns, making them impossible to distinguish based on tissue microstructure. (Credit: Jurmeister/Charité)
Pathologists often have trouble distinguishing pulmonary metastases of head and neck squamous cell carcinoma from primary lung squamous cell carcinoma.
But using artificial intelligence (AI), researchers developed a new classification method which identified the primary origins of cancerous tissue based on chemical DNA changes.
The neural network achieved an accuracy over 99% when distinguishing between lung cancer and head and neck cancer, according to the findings of a study published in the journal Science Translational Medicine.
"The bottom line is that we are solving a highly clinically relevant diagnostic problem by combining DNA-methylation profiling and AI and deep neural networks and got 99% accuracy," Klauschen said in a statement to Inside Digital Health™. "Before our method, this was more less guess work."
Patients with head and neck cancer sometimes develop lung cancer. And it is often difficult to determine if the cancers represent pulmonary metastases of the patient’s head and neck cancer or a second primary cancer, said Frederick Klauschen, M.D., Ph.D., a pathology professor at Charité's Institute in Germany.
“This distinction is hugely important in the treatment of people affected by these cancers,” Klauschen said. “While surgery may provide a cure in patients with localized lung cancers, patients with metastatic head and neck cancers fare significantly worse in terms of survival and will require treatments such as chemoradiotherapy.”
Generally, pathologists analyze the cancer’s microstructure and detect characteristic proteins in the tissue to distinguish between metastases and a second primary tumor, according to the researchers. These tests, however, are usually inconclusive due to the similarities between head and neck cancers and lung cancers.
“In order to solve this problem, we tested tissue samples for a specific chemical alteration known as DNA methylation,” said David Capper, M.D., a professor in the neuropathology department at Charité's Institute.
The research team developed three machine-learning models based on artificial neural networks, support vector machines and random forests. Researchers trained the models on DNA methylation data from more than 200 patients with head and neck and lung cancers. The models helped distinguish between the two types of cancer.
The artificial neural network and the support vector machine correctly classified 96.4 and 95.7% of all cases in the validation cohort. The random forest model achieved an accuracy of 87.8%. The areas under the curve for the artificial neural network, support vector machine and random forest classifier were 0.993, 0.991 and 0.97.
The support vector machine (98.1%) achieved the highest positive predictive value for classifying head and neck cancers, while the neural network (96.7%) achieved the highest for classifying lung cancers.
The researchers are testing the implementation of this method in routine practice to ensure patients will benefit from the results of the study as quickly as possible, Klauschen said.
“(AI) is playing an increasingly important role, not only in our daily lives and in industry, but also in natural sciences and medical research,” said Klaus-Robert Müller, Ph.D., professor of machine learning at the Technical University of Berlin in Germany. “The use of (AI) is, however, particularly complex within the medical field; this is why, until now, research findings have only rarely delivered direct benefits for patients. This could now be about to change.”
Get the best insights in digital health directly to your inbox.