Search

OBJECTIVE In this prospective, randomised study a double lumen tube (DLT) or an Arndt bronchial blocker (BB) was used for one lung ventilation (OLV). Intraoperative oxygenation, ventilation, haemodynamics, and ease of placement were compared. METHODS In 32 patients undergoing robotic harvesting of the left internal thoracic artery (ITA) a left-sided DLT or a BB controlled by bronchoscopy was used for OLV of the right lung. In 2 patients randomised for DLT, a single lumen tube and BB was used for difficult visualisation of the larynx. These 2 patients were excluded from the study. RESULTS Correct placement of DLT (8.8 +/- 5.4 min, 87.5 % of the patients) as well as BB (6.5 +/- 4.0 min; n. s.; 100 % of the patients) was achieved rapidly. The Cormack-Lehane score for visualisation of the larynx was 2, 1 - 3 (median, range) in group DLT and 3, 1 - 4 (median, range) in group BB (p < 0.05). OLV lasted for 93 +/- 29 min in group DLT and 88 +/- 34 min in group BB (n. s.). Satisfactory conditions for harvesting ITA were achieved in all patients of both groups. There was no difference in any haemodynamic parameter at any time between the two groups. Mean pulmonary arterial pressure and pulmonary vascular resistance were increased without significant differences between the two groups during thoracoscopy for robotic preparation of left ITA. During OLV, index of paO (2)/FiO (2) was significantly (p < 0.05) lower in DLT patients compared to BB patients. Time to extubation did not differ between the two groups (DLT 14.3 +/- 4.6 h; BB 11.8 +/- 3.8 h). CONCLUSION OLV in patients undergoing robotic cardiac surgery was safely achieved either by BB or by left-sided DLT. Positioning of the BB was easier than that of a DLT when intubation was difficult and tube exchange at the end of surgery is avoided. Therefore, we prefer a BB in patients undergoing robotic harvesting of left ITA.

Levosimendan, is a new calcium sensitiser with 2 major effects. First, levosimendan acts as a positive inotropic agent by binding calcium dependently to cardiac troponin C. Second, levosimendan activates adenosine triphosphate-regulated potassium (K (ATP)) channels. Thus it has vasodilatory properties and cardioprotective effects at a dose enhancing myocardial contractility. These unique properties of levosimendan might be of great advantage in patients with myocardial ischemia simultaneously requiring inotropic support. The concept of perioperative inoprotection is presented in 6 patients with acute ischemia undergoing emergent cardiac surgery.

This work proposes a novel general framework, in the context of eXplainable Artificial Intelligence (XAI), to construct explanations for the behaviour of Machine Learning (ML) models in terms of middle-level features which represent perceptually salient input parts. One can isolate two different ways to provide explanations in the context of XAI: low and middle-level explanations. Middle-level explanations have been introduced for alleviating some deficiencies of low-level explanations such as, in the context of image classification, the fact that human users are left with a significant interpretive burden: starting from low-level explanations, one has to identify properties of the overall input that are perceptually salient for the human visual system. However, a general approach to correctly evaluate the elements of middle-level explanations with respect ML model responses has never been proposed in the literature. We experimentally evaluate the proposed approach to explain the decisions made by an Imagenet pre-trained VGG16 model on STL-10 images and by a customised model trained on the JAFFE dataset, using two different computational definitions of middle-level features and compare it with two different XAI middle-level methods. The results show that our approach can be used successfully in different computational definitions of middle-level explanations.

In neural networks literature, there is a strong interest in identifying and defining activation functions which can improve neural network performance. In recent years there has been a renovated interest of the scientific community in investigating activation functions which can be trained during the learning process, usually referred to as "trainable", "learnable" or "adaptable" activation functions. They appear to lead to better network performance. Diverse and heterogeneous models of trainable activation function have been proposed in the literature. In this paper, we present a survey of these models. Starting from a discussion on the use of the term "activation function" in literature, we propose a taxonomy of trainable activation functions, highlight common and distinctive proprieties of recent and past models, and discuss main advantages and limitations of this type of approach. We show that many of the proposed approaches are equivalent to adding neuron layers which use fixed (non-trainable) activation functions and some simple local rule that constraints the corresponding weight layers., Published in "Neural Networks" journal (Elsevier)

Automatically learning the best activation function for the task is an active topic in neural network research. At the moment, despite promising results, it is still challenging to determine a method for learning an activation function that is, at the same time, theoretically simple and easy to implement. Moreover, most of the methods proposed so far introduce new parameters or adopt different learning techniques. In this work, we propose a simple method to obtain a trained activation function which adds to the neural network local sub-networks with a small number of neurons. Experiments show that this approach could lead to better results than using a pre-defined activation function, without introducing the need to learn a large number of additional parameters.

Several projects and already-operative observatories aimed at detecting High Energy Cosmic Rays (HECR) are/will be equipped with instruments to monitor the atmosphere. Since cloud presence can affect the night-time indirect measurements of the HECRs and Cherenkov radiation, it is crucial to know the meteorological conditions during the observation period of the HECR detectors. Several meteorological satellites already provide useful information, however to obtain accurate reconstructions of the detected events it is more suitable using devices that operate synchronously with the main detector. To this purpose, infrared cameras that acquire images of the whole field of view are thought to support the atmosphere monitoring during observations from both space and ground. Meaningful parameters, like cloud coverage and cloud top/bottom height, can be retrieved from the analysis of those data. Multi-spectral information are typically analyzed and combined to obtain cloud masks for each image, where a cloudy/cloud-free probability flag is associated with each pixel. These algorithms normally use several spectral bands that are not always available in non-meteorological sensors. A different approach is presented in this paper. It only relies on the gray level values of the image pixel, and it can be applied on thermal infrared as well as visible images acquired from both space and ground. To test the method on real cloudy scenes, images from polar satellite and all-sky data archives are considered, and the results are compared to the corresponding cloudiness masks provided by the same data repositories.

Providing algorithmic explanations for the decisions of machine learning systems to end users, data protection officers, and other stakeholders in the design, production, commercialisation and use of machine learning systems pipeline is an important and challenging research problem. Much work in this area focuses on image classification, where the required explanations can be given in terms of images, therefore making explanations relatively easy to communicate to end-users. For a classification problem, a contrastive explanation tries to understand why the classifier has not answered a particular class, say B, instead of the returned class A. Sparse dictionaries have been recently used to identify local image properties as main ingredients for a system producing humanly understandable explanations for the decisions of a classifier developed based on machine learning methods. In this paper, we show how the system mentioned above can be extended to produce contrastive explanations.

This paper presents an ongoing study for the estimation of the cloud-top height by using only geometrical methods. In agreement with some recent studies showing that it is possible to achieve reliable height estimations not only with the classical methods based on radiative transfer, this article includes a comparison of performances of a selected set of vision algorithms devoted to extract dense disparity maps or motion fields from Infra Red stereo image pairs. This collection includes both area-based techniques and an optical flow-based method and the comparison is accomplished by using a set of cloudy scenes selected from the Along-Track Scanning Radiometer (ATSR2) database. The first group of algorithms yields results generally comparable in terms of the measures used for the inter-comparison, while the maps from the optical flow-based method are slightly better.

This paper presents an ongoing study for the estimation of the cloud-top height by using only geometrical methods. It is based on the hypothesis that an infra-red camera is on board a satellite and pairs of images concern nearly the same scene. Stereo-vision techniques are therefore explored in order to test the methodology for height retrieval and in particular results of several techniques of stereo matching are evaluated. This study includes area-based matching algorithms by implementing the basic versions, without considering any further steps of optimisation to improve the results. Dense depth maps are the final outputs whose reliability is verified by computing error statistics with respect to a set of Digital Terrain Elevation Data, used as ground truth for a set of nearly cloud free images. A set of real pairs of images from the Along-Track Scanning Radiometer2 11μm data set, has been considered. The evaluated errors range between .75 and .80 km, that is not a particularly bad result if it is compared to the resolution of the ATSR2 pixel (1 km resolution).

In this paper we present some preliminary results on an opticalfow based technique aimed at recovering the cloud-top height from infra-red image sequences. The recovery of the cloud-top height from satellite infra-red images is an important topic in meteorological studies, and is traditionally based on the analysis of the temperature maps. In this work we explore the feasibility for this problem of a technique based on a robust multi-resolution opticalfow algorithm. The robustness is achieved adopting a Least Median of Squares paradigm. The algorithm has been tested on semi-synthetic data (i.e. real data that have been synthetically warped in order to have a reliable ground truth for the motion eld), and on real short sequences (pairs of frames) coming from the ATSR2 data set. Since we assumed the same geometry as for the ATSR2 data, the cloud top height could be recovered from the motion eld by means of the widely used Prata and Turner equation.

The paper introduces a software system for detecting and tracking starfish in an underwater video sequence. The target of such a system is to help biologists in giving an estimate of the number of starfish present in a particular area of the sea-bottom. The nature of the input images is characterised by a low signal/noise ratio and by the presence of noisy background represented by pebbles; this makes the detection a non-trivial task. The procedure we use is a chain of several steps that starts from the extraction of the area of interest and ends with a classifier and a tracker providing the necessary information for counting the starfish present in the scene. © 2003 IEEE.

Iconic programming intends to provide expressive tools to implement, to debug, and to execute programs. For this purpose, visual languages need pictorial constructs and metaphors to guide the design of algorithms in interactive fashion. In the paper a new class of dynamic visual interfaces, named DIVA (Dynamic Interface for Visual Applications), is introduced, its properties are described, and an application to visual compilers in a multi-processors system dedicated to image analysis is given. Moreover, a formal definition of dynamic icon (DI) is also given.

Algorithms, based on information fusion, are often embodied in visual perception systems. Distributed architectures have been recently proposed to perform integrated computation. The complexity of distributed systems regards both their design, and the software environment to develop applications. Visual and iconic programming style intends to provide expressive tools to implement, to debug, and to execute programs in distributed environment. Multi-layers graphs languages seem suitable to handle such complexity. This paper describes the design of a visual dynamic environment (VDE), which is based on a graph-grammar. A new class of dynamic visual interfaces is also introduced, and its properties are described. The proposed VDE has been implemented on the first emulated version of the machine M-VIF (Machine Vision based on Information Fusion).