Artificial-intelligence technology that examines images of lung tissue can identify two common lung cancers just as well as pathologists do.
A team led by Narges Razavian and Aristotelis Tsirigos at the New York University School of Medicine trained and tested a convolutional neural network — a deep-learning algorithm that is adept at processing images — on 1,634 images of cancerous and healthy lung tissue. The algorithm identified healthy cases and distinguished as accurately as three pathologists between two common types of lung cancer: adenocarcinoma and squamous-cell carcinoma.
The researchers also trained the network on adenocarcinoma images that had been labelled with the mutations underlying the cancer. After this training, the algorithm was able to accurately predict the mutations associated with some unlabelled images.
If the algorithm were trained on further labelled images, it might be able to identify adenocarcinomas’ mutations with greater accuracy, the researchers say. That could improve this tumour’s treatment, which is often tailored to the underlying mutation.