A Roadmap for Foundational Research on Artificial Intelligence in Medical Imaging: From the 2018 NIH/RSNA/ACR/The Academy Workshop.

Imaging research laboratories are rapidly creating machine learning systems that achieve expert human performance using open-source methods and tools. These artificial intelligence systems are being developed to improve medical image reconstruction, noise reduction, quality assurance, triage, segmentation, computer-aided detection, computer-aided classification, and radiogenomics. In August 2018, a meeting was held in Bethesda, Maryland, at the National Institutes of Health to discuss the current state of the art and knowledge gaps and to develop a roadmap for future research initiatives. Key research priorities include: 1, new image reconstruction methods that efficiently produce images suitable for human interpretation from source data; 2, automated image labeling and annotation methods, including information extraction from the imaging report, electronic phenotyping, and prospective structured image reporting; 3, new machine learning methods for clinical imaging data, such as tailored, pretrained model architectures, and federated machine learning methods; 4, machine learning methods that can explain the advice they provide to human users (so-called explainable artificial intelligence); and 5, validated methods for image de-identification and data sharing to facilitate wide availability of clinical imaging data sets. This research roadmap is intended to identify and prioritize these needs for academic research laboratories, funding agencies, professional societies, and industry.

Trade-offs among catch, bycatch, and landed value in the American Samoa longline fishery.

The interspecific preferences of fishes for different depths and habitats suggest fishers could avoid unwanted catches of some species while still effectively targeting other species. In pelagic longline fisheries, albacore (Thunnus alalunga) are often caught in relatively cooler, deeper water (>100 m) than many species of conservation concern (e.g., sea turtles, billfishes, and some sharks) that are caught in shallower water (<100 m). From 2007 to 2011, we examined the depth distributions of hooks for 1154 longline sets (3,406,946 hooks) and recorded captures by hook position on 2642 sets (7,829,498 hooks) in the American Samoa longline fishery. Twenty-three percent of hooks had a settled depth <100 m. Individuals captured in the 3 shallowest hook positions accounted for 18.3% of all bycatch. We analyzed hypothetical impacts for 25 of the most abundant species caught in the fishery by eliminating the 3 shallowest hook positions under scenarios with and without redistribution of these hooks to deeper depths. Distributions varied by species: 45.5% (n = 10) of green sea turtle (Chelonia mydas), 59.5% (n = 626) of shortbill spearfish (Tetrapturus angustirostris), 37.3% (n = 435) of silky shark (Carcharhinus falciformis), and 42.6% (n = 150) of oceanic whitetip shark (C. longimanus) were caught on the 3 shallowest hooks. Eleven percent (n = 20,435) of all tuna and 8.5% (n = 10,374) of albacore were caught on the 3 shallowest hooks. Hook elimination reduced landed value by 1.6-9.2%, and redistribution of hooks increased average annual landed value relative to the status quo by 5-11.7%. Based on these scenarios, redistribution of hooks to deeper depths may provide an economically feasible modification to longline gear that could substantially reduce bycatch for a suite of vulnerable species. Our results suggest that this method may be applicable to deep-set pelagic longline fisheries worldwide.