The ethical adoption of artificial intelligence in radiology.

Artificial intelligence (AI) is rapidly transforming healthcare-with radiology at the pioneering forefront. To be trustfully adopted, AI needs to be lawful, ethical and robust. This article covers the different aspects of a safe and sustainable deployment of AI in radiology during: training, integration and regulation. For training, data must be appropriately valued, and deals with AI companies must be centralized. Companies must clearly define anonymization and consent, and patients must be well-informed about their data usage. Data fed into algorithms must be made AI-ready by refining, purification, digitization and centralization. Finally, data must represent various demographics. AI needs to be safely integrated with radiologists-in-the-loop: guiding forming concepts of AI solutions and supervising training and feedback. To be well-regulated, AI systems must be approved by a health authority and agreements must be made upon liability for errors, roles of supervised and unsupervised AI and fair workforce distribution (between AI and radiologists), with a renewal of policy at regular intervals. Any errors made must have a root-cause analysis, with outcomes fedback to companies to close the loop-thus enabling a dynamic best prediction system. In the distant future, AI may act autonomously with little human supervision. Ethical training and integration can ensure a "transparent" technology that will allow insight: helping us reflect on our current understanding of imaging interpretation and fill knowledge gaps, eventually moulding radiological practice. This article proposes recommendations for ethical practise that can guide a nationalized framework to build a sustainable and transparent system.

Heterogeneity of pancreatic cancer metastases in a single patient revealed by quantitative proteomics.

Many patients with pancreatic cancer have metastases to distant organs at the time of initial presentation. Recent studies examining the evolution of pancreatic cancer at the genetic level have shown that clonal complexity of metastatic pancreatic cancer is already initiated within primary tumors, and organ-specific metastases are derived from different subclones. However, we do not yet understand to what extent the evolution of pancreatic cancer contributes to proteomic and signaling alterations. We hypothesized that genetic heterogeneity of metastatic pancreatic cancer results in heterogeneity at the proteome level. To address this, we employed a model system in which cells isolated from three sites of metastasis (liver, lung, and peritoneum) from a single patient were compared. We used a SILAC-based accurate quantitative proteomic strategy combined with high-resolution mass spectrometry to analyze the total proteome and tyrosine phosphoproteome of each of the distal metastases. Our data revealed distinct patterns of both overall proteome expression and tyrosine kinase activities across the three different metastatic lesions. This heterogeneity was significant because it led to differential sensitivity of the neoplastic cells to small molecule inhibitors targeting various kinases and other pathways. For example, R428, a tyrosine kinase inhibitor that targets Axl receptor tyrosine kinase, was able to inhibit cells derived from lung and liver metastases much more effectively than cells from the peritoneal metastasis. Finally, we confirmed that administration of R428 in mice bearing xenografts of cells derived from the three different metastatic sites significantly diminished tumors formed from liver- and lung-metastasis-derived cell lines as compared with tumors derived from the peritoneal metastasis cell line. Overall, our data provide proof-of-principle support that personalized therapy of multiple organ metastases in a single patient should involve the administration of a combination of agents, with each agent targeted to the features of different subclones.

A draft map of the human proteome.

The availability of human genome sequence has transformed biomedical research over the past decade. However, an equivalent map for the human proteome with direct measurements of proteins and peptides does not exist yet. Here we present a draft map of the human proteome using high-resolution Fourier-transform mass spectrometry. In-depth proteomic profiling of 30 histologically normal human samples, including 17 adult tissues, 7 fetal tissues and 6 purified primary haematopoietic cells, resulted in identification of proteins encoded by 17,294 genes accounting for approximately 84% of the total annotated protein-coding genes in humans. A unique and comprehensive strategy for proteogenomic analysis enabled us to discover a number of novel protein-coding regions, which includes translated pseudogenes, non-coding RNAs and upstream open reading frames. This large human proteome catalogue (available as an interactive web-based resource at http://www.humanproteomemap.org) will complement available human genome and transcriptome data to accelerate biomedical research in health and disease.