In a recent study published in Nature Communications, researchers from NYU Langone Health's Perlmutter Cancer Center and the University of Glasgow have developed a cutting-edge AI program capable of accurately diagnosing adenocarcinoma, the most common form of lung cancer.
This development, which uses advanced image analysis techniques to improve patient outcomes, represents a significant advancement in cancer diagnosis and management.
(Photo : National Cancer Institute on Unsplash) AI program accurately diagnoses lung cancer and predicts recurrence with 72% accuracy, surpassing human pathologists.
Breakthrough in Lung Diagnosis and Treatment
Diagnosing and managing lung cancer traditionally involves pathologists meticulously examining tissue samples under a microscope. This process is not only time-consuming but also susceptible to human error and bias.
Existing AI methods using supervised deep learning have shown promise but require vast amounts of annotated data, which are costly and labor-intensive to produce.
The new AI program utilizes a technique called Histomorphological Phenotype Learning (HPL), which employs self-supervised learning. Unlike previous methods, this approach does not require pre-labeled data.
Instead, the AI automatically identifies and groups similar-looking areas, or "tiles," within tissue images. These grouped areas form an HP-Atlas, a detailed map showcasing different tissue structures and their transitions from benign to malignant states, including intermediate stages marked by inflammation and other reactive changes.
To validate their method, researchers analyzed nearly half a million tissue images from 452 patients with adenocarcinoma.
The results were impressive: the AI program accurately distinguished between adenocarcinoma and squamous cell carcinoma in 99% of cases and predicted cancer recurrence with 72% accuracy. In comparison, pathologists examining the same images achieved a 64% accuracy rate.
The AI program's ability to quickly analyze lung tissue samples and provide accurate predictions represents a significant advancement in cancer care.
"Our computer program can now analyze lung tissue samples in minutes to provide fairly accurate predictions of whether a patient's cancer will return, predictions that exceed the current standard of care for making a prognosis in lung adenocarcinoma," noted Aristotelis Tsirigos, Ph.D., co-senior investigator of the study.
The program not only offers a detailed breakdown of the tissue's content but also assigns each patient a score reflecting their statistical chance of survival and tumor recurrence over up to five years. Researchers emphasize that as more data is added, the AI will become increasingly accurate, and they plan to make the tool freely available after further testing.
The success of this AI program in lung cancer diagnosis opens the door for similar applications in other types of cancer, such as breast, ovarian, and colorectal cancers.
The team also plans to enhance the current program by incorporating additional data from hospital electronic health records and socioeconomic factors, further improving its accuracy and reliability.
In addition to the AI program for adenocarcinoma, other advancements in AI are making significant impacts on lung cancer diagnosis and treatment. For instance, a 2022 study led by Dr. Anil Vachani at the University of Pennsylvania demonstrated the effectiveness of an AI-based system in assessing cancer risk in lung nodules visible on CT scans.
