Site Editor
Sandy Srinivas, MD
Sandy Srinivas, MD
Posted: Tuesday, August 30, 2022
As reported in NPJ Digital Medicine, a team of researchers assessed scalable tools to support prostate cancer personalization. Using multimodal deep learning architecture, the researchers trained models to predict long-term, clinically relevant outcomes. Osama Mohamad, MD, PhD, of the University of California San Francisco, and colleagues identified a multimodal artificial intelligence (AI) system with the potential to serve as a digital biomarker for global adoption.
Using histopathologic data from more than 5,600 patients from across hundreds of clinical centers, the team trained and validated prognostic biomarkers in localized prostate cancer to develop personalized care. The median follow-up period was 11.4 years. The model learns from and is trained using some of the most accurate clinical and outcomes data available, including both tabular clinical and image data from five phase III randomized trials. The researchers measured the performance of the model with the area under the time–dependent receiver operator characteristic curve of sensitivity and specificity.
Compared with most common risk-stratification tools, the researchers’ model reportedly has a stronger discriminatory performance across all endpoints, ranging from 9.2% to 14.6% relative improvement in a held-out validation set. The technique learns from patient-level data and unannotated slides to predict long-term outcomes. According to the researchers, this self-supervised learning may lower barriers for physicians and clinics to use this tool and easily share data to predict patient outcomes and aid in cancer care decision-making.
“This artificial intelligence–based tool improves prognostication over standard tools and allows oncologists to computationally predict the likeliest outcomes of specific patients to determine optimal treatment. Outfitted with digital scanners and Internet access, any clinic could offer such capabilities, enabling global access to therapy personalization,” the authors commented.
Disclosures: Full authors’ disclosures are available at www.nature.com.
JNCCN Spotlights
Resources
For your Patients
