A team of researchers at NYU School of Medicine have developed a new computer program that can analyze images of patients’ lung tumors and even identify altered genes that drive abnormal cell growth. The machine learning program could distinguish with 97 percent accuracy between adenocarcinoma and squamous cell carcinoma, two types of lung cancer that troubled pathologists sometimes as they require confirmatory tests.
The AI tool was able to determine whether abnormal versions of 6 genes linked to lung cancer, which include EGFR, KRAS, and TP53, are present in the cell or not. Such genetic changes or mutations can cause abnormal growth often seen in cancer. Also, it can change a cell’s shape and interaction with its surroundings while providing visual clues for thorough analyses.
About 20 percent of patients with adenocarcinoma have mutations in the gene epidermal growth factor receptor or EGFR. Determining which genes are changes in which tumor has become vital with the increased use of targeted therapies that work only for specific mutations. The genetic tests currently used to confirm the presence of mutations can take a week to give out results.
In the study, the research team developed statistical techniques that enabled the program to learn how to get better at operations. Such programs create rules and mathematical models that allow decision-making based on data examples fed into them, with the program getting better as the amount of training data grows.
New and better AI approaches, inspired by nerve cell networks in the brain use highly complex circuits to process information, with each step feeding information and assisting with operations all the way.
The current team trained the deep convolutional neural network, Google ‘s Inception v3, to examine slide images obtained from The Cancer Genome Atlas, which is a database with images of cancer diagnoses that have already been examined. This lets the researchers measure their program’s efficacy and automatically classify normal and diseases tissue.
Surprisingly, the study found that more than half of the small percentage of tumor images were misclassified by the study’s AI program were also misclassified by the pathologists. This highlighted the difficulty in distinguishing between two lung cancer types. 45 out of 54 images misclassified by pathologists were assigned to the correct cancer type by the AI program. This shows significant promise regarding the efficacy of the tool and AI could offer a useful second opinion.
“In our study, we were excited to improve on pathologist-level accuracies, and to show that AI can discover previously unknown patterns in the visible features of cancer cells and the tissues around them,” says co-corresponding author Narges Razavian, Ph.D., assistant professor in the departments of Radiology and Population Health. “The synergy between data and computational power is creating unprecedented opportunities to improve both the practice and the science of medicine.”
The team plans to keep training the AI program with data until it can distinguish which genes are mutated in a given cancer with more than 90 percent accuracy, at which point they will start seeking government approval to apply the technology clinically, a and in the diagnosis of various cancer types.