The Frontiers of Artificial Intelligence

D learning is perhaps the most powerful tool driving commercial applications in artificial intelligence (AI) today. This type of advanced computing can grind through gigantic volumes of data, teaching itself how to detect and classify patterns or anomalies and make predictions and recommendations. Deep-learning models can recognize more subtle, complex features faster and often more accurately than people can manage on their own. Take Fimmic Oy, a Finnish spinoff company from the University of Helsinki, which created the WebMicroscope cloud platform. It is the first commercial tool integrating deep learning and computer vision, another AI tool, to help pathologists and researchers find, classify, and grade elusive cancers in tissues. “Deep learning has the potential to change the whole field of medical-image diagnostics,” says Kaisa Helminen, Fimmic’s CEO. Pathologists typically assess tissue slides or images by eye. But Fimmic developed the WebMicroscope Deep Learning AI, which can “see” what’s on a tissue image automatically. The model locates, detects, measures, characterizes features of, and grades tumors on the basis of their morphology. “Pathologists don’t have to estimate where the tumor is, how many stained cells are there, and the tumor’s grade,” says Helminen. “We can give all that information to them before they even start looking at the digital image.” After that, the pathologist verifies the result and makes the final scoring. Research institutions and pharmaceutical companies are working with WebMicroscope in preclinical investigations. It is in the pilot or proof-ofconcept phase with many customers globally. Helminen anticipates a future in clinical use that is not far away. Major corporations are now spending huge sums on deep-learning models for a variety of healthcare applications. Other large companies are partnering with startup software firms to use deep-learning networks in imagery challenges, which could eventually allow underserved patients to receive diagnoses from medical specialists. Computer scientists and botanists are using deep learning to identify plant species from natural-history collections, aiding underfunded taxonomists. Companies are developing deep-learning tools and target applications that can improve yields by diagnosing crop disease and helping farmers fight pests and weeds. For researchers, these tools will probably change how many workers in biology-related disciplines do their jobs.