AI’s ability to outperform human lung cancer detection is rapidly evolving, as was highlighted in the clinical science symposium, Using “Artificial” Intelligence to Achieve “Real” Improvements in Cancer Care, at the 2024 ASCO Annual Meeting.
In that session, Chiharu Sako, PhD, and colleagues reported that they developed and externally validated a generalizable computed tomography (CT) imaging-based biomarker to predict response to immune checkpoint inhibitor (ICI) therapy. Their deep learning radiomic biomarker used a real-world dataset (RWD) of 1,188 stage IV non-small cell lung cancer (NSCLC) patients who were treated with PD-(L)1 ICIs in academic and community sites in the US and Europe.
ICI therapy is standard-of-care for mutation-negative advanced NSCLC treatment, but PD-(L)1 markers are known to be inaccurate, creating the need to identify the patients most likely to benefit from ICI.
“In our validations in RWD and clinical trial cohorts, a deep-learning radiomic biomarker based on routine pre-treatment CT scans predicted response to ICI and stratified patients independently from PD-L1 status,” the authors wrote in their abstract. “This tool may inform clinical decision-making, such as to help guide whether concomitant chemotherapy may not be needed,” the authors concluded. “In future work, we plan to further validate our approach in larger prospective datasets and expand its use to new indications.”
More Commercial Tools and Research
Commercial and academic research, development, and production of AI tools to expedite lung cancer detection and treatment is growing, with recent innovations including:
At the 2024 ASCO Annual Meeting, Qure.ai Technologies Limited demonstrated its proprietary AI-powered solutions to identify, measure, manage, and monitor end-to-end lung cancer patient care. According to a press release, the company’s technology identifies missed lung nodules from chest X-rays, measures lung nodules in CT scans and tracks volumetric growth, manages lung cancer cases, and monitors drug efficacy and disease progression in clinical trials and other studies.
Optellum Virtual Nodule Clinic software uses multiple databases of CT lung scans and diagnoses from various healthcare systems in the United States, the United Kingdom, and Europe to optimize the care of patients with suspected lung cancer. Its AI differentiates between malignant and benign nodules at stage 1A. The system, which was demonstrated at the 2024 ASCO Annual Meeting, applies neural network analytics and standard CT scans to generate lung cancer prediction scores from 1 to 10, with 10 indicating the highest risk of malignancy for each nodule in seconds.
Lucem Health’s Reveal for Lung Cancer uses existing electronic health record (EHR) data and its AI algorithm to help healthcare organizations better identify and more effectively engage patients who are significantly more likely to be diagnosed with lung cancer by standard low-dose CT screening, including ever-smokers 40-89 years of age who are at higher risk for lung cancer.
Researchers at the Johns Hopkins Kimmel Cancer Center and other institutions used AI to potentially identify DNA fragment patterns associated with lung cancer, and they developed and validated a liquid biopsy to potentially identify people more likely to have lung cancer.
As they wrote in Cancer Discovery, coauthor Victor E. Velculescu, MD, PhD, and colleagues prospectively enrolled around 1,000 participants with and without cancer who met the criteria for traditional low-dose CT lung cancer screening at 47 study sites in the United States. They identified those who were most at risk and would benefit from follow-up CT screening.
Regina Barzilay, PhD, of the Massachusetts Institute of Technology, and colleagues developed Sybil, an AI tool to assess lung cancer risk and published their results in Journal of Clinical Oncology. The researchers trained Sybil on hundreds of CT scans with visible cancerous tumors and then tested the model on CT scans without such signs.
Sybil analyzes low-dose computed tomography (LDCT) image data without a radiologist to predict the risk of developing future lung cancer within six years. Sybil obtained C-indices of 0.75, 0.81, and 0.80 over six years from diverse sets of lung LDCT scans, with C-index scores over 0.7 considered good and those over 0.8 considered strong. The ROC-AUCs for one-year prediction using Sybil scored between 0.86 and 0.94.
As reported in Nature Communications, researchers at NYU Langone Health and the University of Glasgow developed and tested an AI-powered computer program based on data from almost 500,000 tissue images that can accurately diagnose adenocarcinoma. According to Aristotelis Tsirigos, PhD, and coauthors, because the program incorporates structural features of tumors from 452 adenocarcinoma patients in the National Cancer Institute’s Cancer Genome Atlas, it provides a detailed, unbiased, reliable second opinion about the presence of the cancer and its prognosis.
The program determined which structural features were linked with disease severity and tumor recurrence. The histomorphological phenotype learning (HPL) study algorithm showed 99% accuracy in distinguishing between adenocarcinoma and squamous cell cancer. HPL was also 72% accurate at predicting prognosis after therapy vs. 64% accuracy of pathologists who directly examined the same patients’ tumor images.
