Diagnosing head and neck cancers is challenging. It is particularly difficult to distinguish pulmonary metastases of head and neck squamous cell carcinoma (HNSC) from primary lung squamous cell carcinomas (LUSCs). However, differentiating these from each another can have important clinical implications. Now, a team of researchers has developed a machine learning algorithm, based on differential DNA methylation, to distinguish primary lung squamous cell carcinomas (LUSCs) from head and neck metastases. The AI method was able to discriminate between the two types with high accuracy, suggesting its potential as a clinical diagnostic tool.
The research, performed by several groups in Berlin and Heidelberg, Germany, is published in Science Translational Medicine in a paper entitled, “Machine learning analysis of DNA methylation profiles distinguishes primary lung squamous cell carcinomas from head and neck metastases.”
Every year, more than 17,000 people in Germany are diagnosed with head and neck cancers. These include cancers of the oral cavity, larynx and nose, but can also affect other areas of the head and neck. Some head and neck cancer patients will also develop lung cancer.
“In the large majority of cases, it is impossible to determine whether these represent pulmonary metastases of the patient’s head and neck cancer or a second primary cancer, i.e. primary lung cancer,” explains Frederick Klauschen, MD, co-senior author of the study.
“This distinction is hugely important in the treatment of people affected by these cancers,” emphasizes Klauschen. “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.”
When trying to distinguish between metastases and a second primary tumor, pathologists use established techniques such as analyzing the cancer’s microstructure and detecting characteristic proteins in the tissue. However, due to the marked similarities between head and neck cancers and lung cancers in this regard, these tests are usually inconclusive.
Based on previous studies illustrating that DNA methylation patterns in cancer cells are highly dependent on the organ in which the cancer originated, the team tested the tissue samples for DNA methylation.
The group employed AI-based methods to analyze DNA methylation data from several hundred head and neck and lung cancers in order to train a deep neural network to distinguish between the two types of cancer.
They performed DNA methylation profiling of primary tumors and trained three different machine learning methods to distinguish metastatic HNSC from primary LUSC. Their artificial neural network correctly classified 96.4% of the cases in a validation cohort of 279 patients with HNSC and LUSC as well as normal lung controls.
The authors write that “As independent clinical validation of the approach, we analyzed a series of 51 patients with a history of HNSC and a second lung tumor, demonstrating the correct classifications based on clinicopathological properties” adding that “our approach may facilitate the reliable diagnostic differentiation of pulmonary metastases of HNSC from primary LUSC to guide therapeutic decisions.”
Klauschen notes that, “to ensure that patients with head and neck cancers and additional lung cancers will benefit from the results of our study as quickly as possible, we are currently in the process of testing the implementation of this diagnostic method in routine practice. This will include a prospective validation study to ensure that the new method can be made available to all affected patients.”
Klaus-Robert Müller, Ph.D. in theoretical computer science and author on the paper notes that, “Artificial intelligence is playing an increasingly important role, not only in our daily lives and in industry, but also in natural sciences and medical research. The use of artificial intelligence 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.”