Researchers led by Northwestern University's Rodrigo Ferrer-Chavez developed a framework called Physics-based Learning for Imaging Planets (FLIP), which combines physics-based telescope modeling with machine-learning optimization to strip starlight from James Webb Space Telescope images, improving exoplanet detection sensitivity by up to 15 times, according to work published in Nature. Directly imaging exoplanets is difficult because the host star's glare can swamp a planet's faint light, and tiny imperfections in telescope optics leave residual glow. Rather than relying on a fixed reference star or a static model, FLIP adjusts for the telescope's optical imperfections and simulates the glare to be removed. In tests on observations from JWST's Near-Infrared Camera (NIRCam), FLIP produced the first image of the known giant planet 51 Eridani b at an infrared wavelength of 4.3 micrometers, which conventional processing had failed to reveal at that wavelength. The team reports sensitivity improvements of two to 15 times compared to standard techniques, and the researchers write that this is the first time such an approach has demonstrated substantially improved sensitivity on real, on-sky coronagraphic imaging. The team plans to reduce the framework's computational cost and adapt it to handle larger data volumes from ground-based observatories.