Artificial Intelligence Accurately Diagnoses Common Lung Cancers
Artificial intelligence (AI), also referred to as machine intelligence, is powered by machines and computers in contrast to intelligence exhibited by human beings.
In recent years, artificial intelligence has accurately provided medical diagnoses and suggested viable treatments. One such success was just recently announced by the New York University Medical Center and published September 17 in Nature Medicine.
Artificial Intelligence: Machine Accuracy Cancer Diagnosis
NYU researchers trained a computer program to read slides of tissue samples to diagnose two of the most common types of lung cancer. The lung cancers are two common types; adenocarcinomas, which make up about 40% of lung cancers, and squamous cell carcinomas, which make up about 25% to 30% of lung cancers. These lung cancers come from different types of cells and require different treatment regimens.
The diagnosis accuracy rate was an astonishing 97%. Compare this to the second most accurate computational method which has an accuracy rate of 83%.
Artificial Intelligence: Study Results
The NYU scientists used a deep learning method originally developed and published by Google. This program uses AI to teach itself to get better at a task, in this case, diagnosis, without being told exactly how.
Thi AI program was trained using more than 1,600 histopathology slides of lung specimens from The Cancer Genome Atlas (TCGA).
These slides of lung tumor specimens were used as a quality control measure to insure that the identity of the tissue was correct.
Researchers then divided each image into thousands of patches into a grid for the AI computer program to analyze. Each patch was analyzed for visual cues, diagnosed by the AI, and then compared to the sample’s original classification. Altogether, researchers had over 1 million patches with which to train the model.
Overall, The program’s accuracy to distinguish adenocarcinoma from squamous cell carcinoma and normal lung cells was equal to that of experienced pathologists. However, the AI analysis was faster; measured in seconds compared to minutes by pathologists.
The program also correctly classified 45 of 54 images that one of three pathologists participating in the study misclassified. This suggests that AI can offer an accurate second opinion.
Artificial Intelligence: Its Role in Treatment and Research
The AI program speed and accuracy suggest that it could also be used during surgery. For example, it can verify whether the surgeon should take another, better biopsy sample for diagnosis.
In addition, the program also identified six of the most common genetic mutations in lung cancer adenocarcinoma. The accuracy rate here ranged from 64% to 86%, depending on the gene.
Currently, the only way humans can detect genetic mutations is by DNA sequencing, which can take up to 2 weeks. Waiting 2 weeks is bad, as lung cancer is often detected late, and progresses rapidly in the later stages.
Dr. Michael Snyder, Ph.D., chair of genetics at Stanford University, says artificial intelligence as the future of diagnosis. “I think we need to shift to using machine learning rather than rely on pathologists alone to do all the work,” he said. “Algorithms won’t replace pathologists, but they will assist them in making classifications. They will reduce the errors that pathologists would otherwise make.”