AI can identify lung cancer types in images, study finds

Dive Brief:

Researchers at New York University School of Medicine said they have successfully used artificial intelligence to distinguish between two types of lung cancer on slide images of tumors, according to a study published in the journal Nature Medicine.
The AI program determined with 97% accuracy which images showed adenocarcinoma and which showed squamous cell carcinoma. The computer program also was able to identify whether genetic mutations were present in cells with 73% to 86% accuracy.
Because results from genetic tests currently can take weeks to return, AI software that can instantly determine cancer subtype and mutational profile could help patients get started sooner on targeted treatments, the researchers said.

Dive Insight:

Advances in machine learning and affordable cloud computing and storage are speeding development of AI tools for diagnostic imaging. Although not yet mainstream, AI software is being used in applications ranging from improving MRI scans to detecting eye diseases and is predicted to play a major role in radiology.

In the NYU study, the researchers trained Google's Inception v3 network to analyze images of patients' lung tumors from the Cancer Genome Atlas database. More than 1,600 slides were classified into adenocarcinoma, squamous cell carcinoma or normal lung tissue.

Distinguishing between the two forms of lung cancer, adenocarcinoma and squamous cell carcinoma, can be tricky even for experienced pathologists, the researchers said. Half of the small percentage of tumor images misclassified by the AI program in the study were also misclassified by the pathologists. However, 45 of 54 images misclassified by at least one pathologist were identified correctly by the machine learning program. The researchers said this suggested AI could be used as a second opinion.

They also trained the network to identify the most commonly mutated genes and found that six — STK11, EGFR, FAT1, SETBP1, KRAS and TP53 — could be detected from pathology images. The AI program was 73% to 86% accurate in determining whether abnormal versions of the six genes linked to lung cancer were present in cells. Such genetic mutations often cause the abnormal growth in cancer and change a cell’s shape, which can be a visual clue for automated analysis.

Identifying which genes are changed enables the use of targeted therapies that work against cancer cells with specific mutations. The NYU researchers said they plan to keep training their AI program until it can identify the mutated genes with more than 90% accuracy. They also plan to seek government approval to use the technology clinically and in the diagnosis of several cancer types.

"Our study provides strong evidence that an AI approach will be able to instantly determine cancer subtype and mutational profile to get patients started on targeted therapies sooner," said study author Aristotelis Tsirigos, associate professor at NYU Langone's Perlmutter Cancer Center in a statement.