New York University (NYU) has published research that shows how AI might be used to assist in lung cancer diagnosis. The study found that a Google-backed AI or "machine learning" program, could distinguish between two types of lung cancer with 97 percent accuracy, all from just an image.
The program analyzed an image of a slice of cancerous tissue which it could then accurately label as either adenocarcinoma or squamous cell carcinoma, two types of lung cancers, that are difficult even for highly experienced pathologists to identify without additional testing. Understanding exactly what type of cancer a patient means more accurate and specific treatment.
AI determines abnormal genes presence
In this case, the adenocarcinoma and squamous cell carcinoma are treated differently. In addition to identifying cancer types, the AI was able to determine whether abnormal versions of 6 genes linked to lung cancer were present in the cells it examined.
The AI had an accuracy that ranged from 73 to 86 percent depending on the gene. These genetic mutations often cause the cell’s shape to change as well as the way it interacts with its surroundings which can help automated analysis visually read the cell's state.
Accurate diagnosis leads to faster treatment
Usually identifying these changes in a patient's genes takes weeks of waiting for test results. "Delaying the start of cancer treatment is never good," says senior study author Aristotelis Tsirigos, Ph.D., associate professor in the Department of Pathology at NYU Langone Perlmutter Cancer Center.
"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." The authors of the study trained the computer program so that it could learn from its mistakes, without ever being told exactly what to do.
Essentially the system gets smarter as it trains itself. The team used a deep convolutional neural network created by Google called Inception v3 to analyze a huge amount of visual data.
AI continues to improve its accuracy
The scientists fed the AI with slides from The Cancer Genome Atlas, a database with images of cancer diagnoses that have already been determined. By using images that already had a prior diagnosis allowed researchers to see the improvements in their program.
When the AI struggled to correctly identify the tumor images, the researchers discovered the same images had been problematic for pathologists and in some cases had also been misidentified. This proves that accurately identifying the difference between different types of cancer cells is incredibly difficult and supports the need for new innovative diagnosis tools.
"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, PhD, 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 research team will continue to train the AI with data until it has at least a 90 percent accuracy.
From there they will look to gain government approval to use the technology clinically, and in the diagnosis of several cancer types. The study is published in Nature Medicine.