To evaluate whether a real-time electronic alert system or “AKI sniffer,” which is based on the RIFLE classification criteria (Risk, Injury and Failure), would have an impact on therapeutic interventions and acute kidney injury progression. Prospective intervention study. Surgical and medical intensive care unit in a tertiary care hospital. A total of 951 patients having in total 1,079 admission episodes were admitted during the study period (prealert control group: 227, alert group: 616, and postalert control group: 236). Three study phases were compared: A 1.5-month prealert control phase in which physicians were blinded for the acute kidney injury sniffer and a 3-month intervention phase with real-time alerting of worsening RIFLE class through the Digital Enhanced Cordless Technology telephone system followed by a second 1.5-month postalert control phase. A total of 2593 acute kidney injury alerts were recorded with a balanced distribution over all study phases. Most acute kidney injury alerts were RIFLE class risk (59.8%) followed by RIFLE class injury (34.1%) and failure (6.1%). A higher percentage of patients in the alert group received therapeutic intervention within 60 mins after the acute kidney injury alert (28.7% in alert group vs. 7.9% and 10.4% in the pre- and postalert control groups, respectively, p μ .001). In the alert group, more patients received fluid therapy (23.0% vs. 4.9% and 9.2%, p μ .01), diuretics (4.2% vs. 2.6% and 0.8%, p μ .001), or vasopressors (3.9% vs. 1.1% and 0.8%, p μ .001). Furthermore, these patients had a shorter time to intervention (p μ .001). A higher proportion of patients in the alert group showed return to a baseline kidney function within 8 hrs after an acute kidney injury alert “from normal to risk” compared with patients in the control group (p = .048). The real-time alerting of every worsening RIFLE class by the acute kidney injury sniffer increased the number and timeliness of early therapeutic interventions. The borderline significant improvement of short-term renal outcome in the RIFLE class risk patients needs to be confirmed in a large multicenter trial. (Crit Care Med 2012; 40:–1170)

The condition of a machine can automatically be identified by creating and classifying features that summarize characteristics of measured signals. Currently, experts, in their respective fields, devise these features based on their knowledge. Hence, the performance and usefulness depends on the expert's knowledge of the underlying physics or statistics. Furthermore, if new and additional conditions should be detectable, experts have to implement new feature extraction methods. To mitigate the drawbacks of feature engineering, a method from the subfield of feature learning, i.e., deep learning (DL), more specifically convolutional neural networks (NNs), is researched in this paper. The objective of this paper is to investigate if and how DL can be applied to infrared thermal (IRT) video to automatically determine the condition of the machine. By applying this method on IRT data in two use cases, i.e., machine-fault detection and oil-level prediction, we show that the proposed system is able to detect many conditions in rotating machinery very accurately (i.e., 95 and 91.67% accuracy for the respective use cases), without requiring any detailed knowledge about the underlying physics, and thus having the potential to significantly simplify condition monitoring using complex sensor data. Furthermore, we show that by using the trained NNs, important regions in the IRT images can be identified related to specific conditions, which can potentially lead to new physical insights.

Hyperspectral image (HSI) classification is one of the most widely used methods for scene analysis from hyperspectral imagery. In the past, many different engineered features have been proposed for the HSI classification problem. In this paper, however, we propose a feature learning approach for hyperspectral image classification based on convolutional neural networks (CNNs). The proposed CNN model is able to learn structured features, roughly resembling different spectral band-pass filters, directly from the hyperspectral input data. Our experimental results, conducted on a commonly-used remote sensing hyperspectral dataset, show that the proposed method provides classification results that are among the state-of-the-art, without using any prior knowledge or engineered features.