Enhancing the reliability and accuracy of AI-enabled diagnosis via complementarity-driven deferral to clinicians.

Predictive artificial intelligence (AI) systems based on deep learning have been shown to achieve expert-level identification of diseases in multiple medical imaging settings, but can make errors in cases accurately diagnosed by clinicians and vice versa. We developed Complementarity-Driven Deferral to Clinical Workflow (CoDoC), a system that can learn to decide between the opinion of a predictive AI model and a clinical workflow. CoDoC enhances accuracy relative to clinician-only or AI-only baselines in clinical workflows that screen for breast cancer or tuberculosis (TB). For breast cancer screening, compared to double reading with arbitration in a screening program in the UK, CoDoC reduced false positives by 25% at the same false-negative rate, while achieving a 66% reduction in clinician workload. For TB triaging, compared to standalone AI and clinical workflows, CoDoC achieved a 5-15% reduction in false positives at the same false-negative rate for three of five commercially available predictive AI systems. To facilitate the deployment of CoDoC in novel futuristic clinical settings, we present results showing that CoDoC's performance gains are sustained across several axes of variation (imaging modality, clinical setting and predictive AI system) and discuss the limitations of our evaluation and where further validation would be needed. We provide an open-source implementation to encourage further research and application.

Proton interaction in phosphate adsorption onto goethite.

The adsorption of phosphate on goethite is generally modeled by assuming a simple ligand exchange reaction with surface hydroxyl groups. This study investigates the binding forms of phosphate on goethite by evaluating the proton interaction and surface charge change during phosphate adsorption. It is found that OH(-) release stoichiometry increases with phosphate coverage, which suggests that different mechanisms predominate at different phosphate loadings. It demonstrates that surface binding changes from monodentate complexation to bidentate complexation with increasing surface phosphate coverage. The net OH(-) release accompanying this transformation is best interpreted with a 2pK(a) multisite model.

Slow adsorption reaction between arsenic species and goethite (alpha-FeOOH): diffusion or heterogeneous surface reaction control.

The slow stage of phosphate or arsenate adsorption on hydrous metal oxides frequently follows an Elovich equation. The equation can be derived by assuming kinetic control by either a diffusion process (either interparticle or intraparticle) or a heterogeneous surface reaction. The aim of this study is to determine whether the slow stage of arsenic adsorption on goethite is more consistent with diffusion or heterogeneous surface reaction control. Adsorption kinetics of arsenate and dimethylarsinate (DMA) on goethite (alpha-FeOOH) were investigated at different pH values and inert electrolyte concentrations. Their adsorption kinetics was described and compared using Elovich (Gamma vs ln time) plots. Desorption of arsenate and DMA was studied by increasing the pH of the suspension from pH 4.0 to pH 10.0 or 12.0. The effective particle sizes and zeta-potential of goethite were also determined. Effective particle size increased rapidly as the pH approached pH(IEP), both in the absence and presence of arsenic. Inert electrolyte concentrations and pH had no effect on the slow stage of arsenate adsorption on goethite, while the kinetics of DMA adsorption on goethite was influenced by both parameters. The slow stage of arsenate adsorption on goethite follows an Elovich equation. Since effective particle size changes with both pH and inert electrolyte concentrations, and effective particle size influences interparticle diffusion, the arsenate adsorption kinetics indicate that the slow adsorption step is not due to interparticle diffusion. DMA also has complex adsorption kinetics with a slow adsorption stage. DMA desorbed completely and rapidly when the pH was raised, in contrast to the slow adsorption kinetics, indicating that the slow adsorption step is not due to intraparticle diffusion. The slow adsorption is not the result of diffusion, but rather is due either to the heterogeneity of the surface site bonding energy or to other reactions controlling arsenic removal from solution.

Evidence for surface precipitation of phosphate on goethite.

Recent studies have suggested that the interaction between phosphate and goethite includes ternary adsorption/ surface precipitation as well as surface complexformation. The ternary adsorption/surface precipitation process envisioned involves the dissolution of the goethite crystal and subsequent adsorption of iron on the surface-bound phosphate. Further evidence to support the suggested process is needed. The process was investigated using two approaches. First, the sorption of iron spiked into a slurry of phosphated goethite and the effect of the iron sorption on phosphate uptake kinetics were investigated to determine whether iron would be adsorbed on the phosphated surface and whether it would enhance phosphate adsorption. Lead was also spiked into solution for comparison. Second, changes in the xi-potential of phosphated goethite were monitored with time. Adsorption of iron on the surface of phosphated goethite should increase the xi-potential of the goethite. Iron spiked into a phosphated goethite slurry was adsorbed on the solid with a concurrent adsorption of phosphate. The iron adsorption did not change the slow phosphate adsorption kinetics. Lead spiked into the solution was also sorbed but to a lesser extent than iron and with a lower apparent P:Pb mole ratio. Lead addition also changed the phosphate adsorption kinetics. With time, the xi-potential of phosphated goethite became more positive, returning almost to the potential of unphosphated goethite at low surface coverages. The slow increase in xi-potential over time indicates that long-term reactions are occurring on the goethite surface, most likely involving the dissolution of goethite to release iron and the subsequent reaction between the iron and surface-bound phosphate. These results provide strong support for the surface precipitation model, and are inconsistent with models envisioning the reaction between phosphate and the goethite surface as involving only monolayer surface complex formation.