In a recent study published in the prestigious journal Nature Communications, researchers from institutions such as New York University and the University of Glasgow have successfully developed an advanced artificial intelligence program capable of accurately diagnosing the most common type of lung cancer—adenocarcinoma—and precisely predicting the risk of recurrence, marking a significant breakthrough in lung cancer diagnosis and treatment.
Traditionally, lung cancer diagnosis requires pathologists to carefully examine tissue samples under a microscope, a process that is not only time-consuming but also prone to human error. Although existing supervised deep learning-based AI methods have shown promising results, they require substantial support from labeled data.
Image source note: The image is generated by AI, authorized by the service provider Midjourney
The AI program in this latest research employs a self-supervised learning technique called "histomorphological phenotype learning." This technology can automatically identify and group similar regions in tissue images, constructing the "HP-Atlas"—a detailed map that shows the transformation process of various tissue structures from benign to malignant states.
The researchers analyzed nearly 500,000 tissue images from 452 adenocarcinoma patients, with encouraging results: the AI program accurately distinguished adenocarcinoma from another common type of lung cancer, squamous cell carcinoma, in 99% of cases, and predicted the risk of tumor recurrence with 72% accuracy, significantly outperforming the 64% accuracy of human diagnosis.
This AI program not only rapidly and comprehensively analyzes lung tissue samples but also generates a score for each patient, accurately reflecting their survival and recurrence probabilities within the next five years. Researchers say that with more data added, the system will become even more accurate and plan to make it available to the public for free after further testing.
Dr. Aristotelis Zolotas from the Perlmutter Cancer Center at New York University said, "Our AI program can analyze lung tissue in minutes and accurately predict whether a patient's cancer will recur, surpassing the current standard requirements for prognosis assessment of lung adenocarcinoma."
This groundbreaking advancement brings more precise and personalized treatment options for lung cancer patients and also opens the door for AI diagnosis in other cancer types, including breast, ovarian, and colorectal cancers. The research team next plans to integrate more clinical and socioeconomic data to further enhance the system's accuracy and reliability.