Deep learning explainability is often reached by gradient-based approaches that attribute the network output to perturbations of the input pixels. However, the relevance of input pixels may be ...difficult to relate to relevant image features in some applications, e.g. diagnostic measures in medical imaging. The framework described in this paper shifts the attribution focus from pixel values to user-defined concepts. By checking if certain diagnostic measures are present in the learned representations, experts can explain and entrust the network output. Being post-hoc, our method does not alter the network training and can be easily plugged into the latest state-of-the-art convolutional networks. This paper presents the main components of the framework for attribution to concepts, in addition to the introduction of a spatial pooling operation on top of the feature maps to obtain a solid interpretability analysis. Furthermore, regularized regression is analyzed as a solution to the regression overfitting in high-dimensionality latent spaces. The versatility of the proposed approach is shown by experiments on two medical applications, namely histopathology and retinopathy, and on one non-medical task, the task of handwritten digit classification. The obtained explanations are in line with clinicians’ guidelines and complementary to widely used visualization tools such as saliency maps.
Display omitted
•Feature attribution explains CNNs in terms of the input pixels.•The abstraction of feature attribution to higher level impacting factors is hard.•Concept attribution explains CNNs with high-level concepts such as clinical factors.•Nuclei pleomorphism is shown as a relevant factor in breast tumor classification.•Concept attribution can match clinical expectations to the interpretability of CNNs.
Protocols for testing conditioned pain modulation (CPM) vary between different labs/clinics. In order to promote research and clinical application of this tool, we summarize the recommendations of ...interested researchers consensus meeting regarding the practice of CPM and report of its results.
Salinity observing satellites have the potential to monitor river fresh-water plumes mesoscale spatio-temporal variations better than any other observing system. In the case of the Soil Moisture and ...Ocean Salinity (SMOS) satellite mission, this capacity was hampered due to the contamination of SMOS data processing by strong land-sea emissivity contrasts. Kolodziejczyk et al. (2016) (hereafter K2016) developed a methodology to mitigate SMOS systematic errors in the vicinity of continents, that greatly improved the quality of the SMOS Sea Surface Salinity (SSS). Here, we find that SSS variability, however, often remained underestimated, such as near major river mouths. We revise the K2016 methodology with: a) a less stringent filtering of measurements in regions with high SSS natural variability (inferred from SMOS measurements) and b) a correction for seasonally-varying latitudinal systematic errors. With this new mitigation, SMOS SSS becomes more consistent with the independent SMAP SSS close to land, for instance capturing consistent spatio-temporal variations of low salinity waters in the Bay of Bengal and Gulf of Mexico. The standard deviation of the differences between SMOS and SMAP weekly SSS is <0.3 pss in most of the open ocean. The standard deviation of the differences between 18-day SMOS SSS and 100-km averaged ship SSS is 0.20 pss (0.24 pss before correction) in the open ocean. Even if this standard deviation of the differences increases closer to land, the larger SSS variability yields a more favorable signal-to-noise ratio, with r2 between SMOS and SMAP SSS larger than 0.8. The correction also reduces systematic biases associated with man-made Radio Frequency Interferences (RFI), although SMOS SSS remains more impacted by RFI than SMAP SSS. This newly-processed dataset will allow the analysis of SSS variability over a larger than 8 years period in regions previously heavily influenced by land-sea contamination, such as the Bay of Bengal or the Gulf of Mexico.
•Improved SMOS salinity systematic error correction from Kolodziejczyk et al. (2016)•Refined variability of sea surface salinity near e.g. major river mouths•Consistent mesoscale patterns observed by SMOS and SMAP satellite missions
The advent of high-throughput sequencing in clinical microbiology is opening the way to new diagnostic and prognostic approaches in infectious diseases. Detection, identification and characterisation ...of pathogenic microorganisms are essential steps in diagnosis and implementation of appropriate antimicrobial therapy. However, standard methods of microbiological diagnosis are failing in some cases. In addition, the emergence of new infections, facilitated by international travel and global warming, requires the implementation of innovative diagnostic methods. Among the different strategies used in clinical microbiology and reviewed in this article, shotgun metagenomics is the only technique that allows today a panpathogenic and unbiased detection of all microorganisms potentially responsible for an infectious disease, including those still unknown. The aims of this article are to present the different possible strategies of high-throughput sequencing used in the microbiological diagnosis of infectious diseases and to highlight the diagnostic contribution of shotgun metagenomics in the field of central nervous system infections.
Colistin pharmacokinetics: the fog is lifting Couet, W.; Grégoire, N.; Marchand, S. ...
Clinical microbiology and infection,
January 2012, 2012-Jan, 2012-01-00, 20120101, 2012-01, Letnik:
18, Številka:
1
Journal Article
Recenzirano
Odprti dostop
Colistin is a re-emerging old antibiotic that is used to treat multidrug-resistant infections in critically ill patients. It corresponds to a mixture of at least 30 different compounds administered ...as inactive derivatives. Therefore, colistin pharmacokinetics are quite difficult to investigate and complex to predict. However specific chromatographic methods have been made available in recent years, leading to a series of modern pharmacokinetic studies after intravenous administration of the prodrug to critical-care patients; these have been conducted by a few groups and have only been recently published. The objective of this article was to conduct a critical review of these very informative modern pharmacokinetic studies and to provide prospective thoughts.
A deficit of endogenous pain inhibitory systems has been suggested to contribute to some chronic pain conditions, one of them being fibromyalgia. The aim of the investigation was to test whether ...endogenous pain inhibitory systems were activated by a spatial summation procedure in 30 fibromyalgia, 30 chronic low back pain, and 30 healthy volunteers who participated in a cross-over trial (two sessions). Each session consisted of visual analog scale ratings of pain during the immersion of different surfaces of the arm in circulating noxious cold (12
°C) water. The arm was arbitrarily divided into eight segments from the fingertips to the shoulder. One session was ascending (from the fingertips to the shoulder) and the other was descending (from the shoulder to the fingertips); they included eight consecutive 2-min immersions separated by 5-min resting periods. For healthy and low back pain subjects, pain was perceived differently during the ascending and descending sessions (
P=0.0001). The descending session resulted in lower pain intensity and unpleasantness. This lowering of the perception curve seems to be due to a full recruitment of inhibitory systems at the beginning of the descending session as opposed to a gradual recruitment during the ascending session. For fibromyalgia subjects, no significant differences were found between the increasing and decreasing sessions (
P>0.05). These data support a deficit of endogenous pain inhibitory systems in fibromyalgia but not in chronic low back pain. The treatments proposed to fibromyalgia patients should aim at stimulating the activity of those endogenous systems.
Colistin pharmacokinetics (PK) was investigated in young healthy volunteers after a 1‐h infusion of 80 mg (1 million international units (MIU)) of the prodrug colistin methanesulfonate (CMS). ...Concentration levels of CMS and colistin were determined in plasma and urine using a new chromatographic assay and analyzed simultaneously with a population approach after correcting the urine‐related data for postexcretion hydrolysis of CMS into colistin. CMS and colistin have low volumes of distribution (14.0 and 12.4 liters, respectively), consistent with distribution being restricted to extracellular fluid. CMS is mainly excreted unchanged in urine (70% on average), with a typical renal clearance estimated at 103 ml/min—close to the glomerular filtration rate. Colistin elimination is essentially extrarenal, given that its renal clearance is 1.9 ml/min, consistent with extensive reabsorption. Colistin elimination is not limited by the formation rate because its half‐life (3 h) is longer than that of CMS. The values of these pharmacokinetic parameters will serve as reference points for future comparisons with patients' data.
Clinical Pharmacology & Therapeutics (2011) 89 6, 875–879. doi:10.1038/clpt.2011.48
Critically ill patients with severe infections are at high risk of suboptimal antimicrobial dosing. The pharmacokinetics (PK) and pharmacodynamics (PD) of antimicrobials in these patients differ ...significantly from the patient groups from whose data the conventional dosing regimens were developed. Use of such regimens often results in inadequate antimicrobial concentrations at the site of infection and is associated with poor patient outcomes. In this article, we describe the potential of in vitro and in vivo infection models, clinical pharmacokinetic data and pharmacokinetic/pharmacodynamic models to guide the design of more effective antimicrobial dosing regimens. Individualised dosing, based on population PK models and patient factors (e.g. renal function and weight) known to influence antimicrobial PK, increases the probability of achieving therapeutic drug exposures while at the same time avoiding toxic concentrations. When therapeutic drug monitoring (TDM) is applied, early dose adaptation to the needs of the individual patient is possible. TDM is likely to be of particular importance for infected critically ill patients, where profound PK changes are present and prompt appropriate antibiotic therapy is crucial. In the light of the continued high mortality rates in critically ill patients with severe infections, a paradigm shift to refined dosing strategies for antimicrobials is warranted to enhance the probability of achieving drug concentrations that increase the likelihood of clinical success.
Background
Omalizumab, an anti‐IgE antibody, is used to treat patients with severe allergic asthma. The evolution of lung function parameters over time and the difference between omalizumab responder ...and nonresponder patients remain inconclusive. The objective of this real‐life study was to compare the changes in forced expiratory volume in 1 second (FEV1) of omalizumab responders and nonresponders at 6 months.
Methods
A multicenter analysis was performed in 10 secondary and tertiary institutions. Lung function parameters (forced vital capacity (FVC), pre‐ and postbronchodilator FEV1, residual volume (RV), and total lung capacity (TLC) were determined at baseline and at 6 months. Omalizumab response was assessed at the 6‐month visit. In the omalizumab responder patients, lung function parameters were also obtained at 12, 18, and 24 months.
Results
Mean prebronchodilator FEV1 showed improvement in responders at 6 months, while a decrease was observed in nonresponders (+0.2±0.4 L and −0.1±0.4 L, respectively, P<.01). After an improvement at 6 months, pre‐ and postbronchodilator FEV1 remained stable at 12, 18, and 24 months. The FEV1/FVC remained unchanged over time, but the proportion of patients with an FEV1/FVC ratio <0.7 decreased at 6, 12, 18, and 24 months (55.2%, 54.0%, 54.0%, and 44.8%, respectively, P<.05). Mean RV values decreased at 6 months but increased at 12 months and 24 months (P<.05). Residual volume/total lung capacity (RV/TLC) ratio decreased at 6 months and remained unchanged at 24 months.
Conclusion
After omalizumab initiation, FEV1 improved at 6 months in responder patients and then remained stable for 2 years. RV and RV/TLC improved at 6 months.