Publications

This paper explores the potential of using the SAM (Segment-Anything Model) segmentator to enhance the segmentation capability of known methods. SAM is a promptable segmentation system that offers zero-shot generalization to unfamiliar objects and images, eliminating the need for additional training. The open-source nature of SAM allows for easy access and implementation. In our experiments, we aim to improve the segmentation performance by providing SAM with checkpoints extracted from the masks produced by mainstream segmentators, and then merging the segmentation masks provided by these two networks. We examine the “oracle” method (as upper bound baseline performance), where segmentation masks are inferred only by SAM with checkpoints extracted from the ground truth. One of the main contributions of this work is the combination (fusion) of the logit segmentation masks produced by the SAM model with the ones provided by specialized segmentation models such as DeepLabv3+ and PVTv2. This combination allows for a consistent improvement in segmentation performance in most of the tested datasets. We exhaustively tested our approach on seven heterogeneous public datasets, obtaining state-of-the-art results in two of them (CAMO and Butterfly) with respect to the current best-performing method with a combination of an ensemble of mainstream segmentator transformers and the SAM segmentator. The results of our study provide valuable insights into the potential of incorporating the SAM segmentator into existing segmentation techniques. We release with this paper the open-source implementation of our method.

The accurate determination of the local temperature is one of the most important challenges in the field of nanotechnology and nanomedicine. For this purpose, different techniques and materials have been extensively studied in order to identify both the best-performing materials and the techniques with greatest sensitivity. In this study, the Raman technique was exploited for the determination of the local temperature as a non-contact technique and titania nanoparticles (NPs) were tested as nanothermometer Raman active material. Biocompatible titania NPs were synthesized following a combination of sol-gel and solvothermal green synthesis approaches, with the aim of obtaining pure anatase samples. In particular, the optimization of three different synthesis protocols allowed materials to be obtained with well-defined crystallite dimensions and good control over the final morphology and dispersibility. TiO2 powders were characterized by X-ray diffraction (XRD) analyses and room-temperature Raman measurements, to confirm that the synthesized samples were single-phase anatase titania, and using SEM measurements, which clearly showed the nanometric dimension of the NPs. Stokes and anti-Stokes Raman measurements were collected, with the excitation laser at 514.5 nm (CW Ar/Kr ion laser), in the temperature range of 293–323 K, a range of interest for biological applications. The power of the laser was carefully chosen in order to avoid possible heating due to the laser irradiation. The data support the possibility of evaluating the local temperature and show that TiO2 NPs possess high sensitivity and low uncertainty in the range of a few degrees as a Raman nanothermometer material.

The Audio Recording Preservation (ARP) technology represents a significant development in the field of audio preservation and is an essential component of the Moving Picture, Audio and Data Coding by Artificial Intelligence (MPAI) Context-based Audio Enhancement (CAE) standard. This standard has been adopted by the IEEE Standard Association as IEEE 3302-2022 and it specifies a range of AI-based technologies for various audio applications, including communication, entertainment, post-production, teleconferencing, and preservation. This article aims to highlight the specific contribution of CAE-ARP technology to audio preservation applications. The CAE-ARP technology has several innovative features that make it a valuable tool in the digitization and preservation of open-reel audio tapes. It leverages automated AI for extracting relevant information from digitized audio files and for creating preservation and access copies. By using the ARP standard, archives can effectively manage all the information stored onto tapes, along with related metadata, to automatically prepare the content for storage and/or immediate use. This technology represents a significant advancement in the field of audio preservation and provides an effective solution for managing small and large collections of open-reel audio tapes.

Background The effect of noninvasive respiratory support (NRS), including high-flow nasal oxygen, bilevel positive airway pressure and continuous positive airway pressure (noninvasive ventilation (NIV)), for preventing and treating post-extubation respiratory failure is still unclear. Our objective was to assess the effects of NRS on post-extubation respiratory failure, defined as re-intubation secondary to post-extubation respiratory failure (primary outcome). Secondary outcomes included the incidence of ventilator-associated pneumonia (VAP), discomfort, intensive care unit (ICU) and hospital mortality, ICU and hospital length of stay (LOS), and time to re-intubation. Subgroup analyses considered “prophylactic” versus “therapeutic” NRS application and subpopulations (high-risk, low-risk, post-surgical and hypoxaemic patients). Methods We undertook a systematic review and network meta-analysis (Research Registry: reviewregistry1435). PubMed, Embase, CENTRAL, Scopus and Web of Science were searched (from inception until 22 June 2022). Randomised controlled trials (RCTs) investigating the use of NRS after extubation in ICU adult patients were included. Results 32 RCTs entered the quantitative analysis (5063 patients). Compared with conventional oxygen therapy, NRS overall reduced re-intubations and VAP (moderate certainty). NIV decreased hospital mortality (moderate certainty), and hospital and ICU LOS (low and very low certainty, respectively), and increased discomfort (moderate certainty). Prophylactic NRS did not prevent extubation failure in low-risk or hypoxaemic patients. Conclusion Prophylactic NRS may reduce the rate of post-extubation respiratory failure in ICU patients.

By leveraging the recent development of artificial intelligence algorithms, several medical sectors have benefited from using automatic segmentation tools from bioimaging to segment anatomical structures. Segmentation of the musculoskeletal system is key for studying alterations in anatomical tissue and supporting medical interventions. The clinical use of such tools requires an understanding of the proper method for interpreting data and evaluating their performance. The current systematic review aims to present the common bottlenecks for musculoskeletal structures analysis (e.g., small sample size, data inhomogeneity) and the related strategies utilized by different authors. A search was performed using the PUBMED database with the following keywords: deep learning, musculoskeletal system, segmentation. A total of 140 articles published up until February 2022 were obtained and analyzed according to the PRISMA framework in terms of anatomical structures, bioimaging techniques, pre/post-processing operations, training/validation/testing subset creation, network architecture, loss functions, performance indicators and so on. Several common trends emerged from this survey; however, the different methods need to be compared and discussed based on each specific case study (anatomical region, medical imaging acquisition setting, study population, etc.). These findings can be used to guide clinicians (as end users) to better understand the potential benefits and limitations of these tools.

The determination of the local temperature is an interesting and intriguing topic in the nanotechnology and nanomedicine world, in terms of tuning the best noninvasive measurement protocol and identification of the more versatile and performing material. In this paper, the Raman technique and titania NPs have been exploited for the realization of a new optical nanotermometer. Biocompatible titania NPs have been properly synthesized, following a combination of sol-gel and solvothermal green synthesis approaches, with the aim of obtaining samples of pure anatase, characterized by crystallite dimensions defined and good control over the final morphology and dispersibility. Powder XRD measurements and room temperature Raman measurements confirmed that the synthesized samples are single-phase anatase. The SEM images clearly showed the nanometric dimension of NPs. Stokes and anti-Stokes Raman measurements, collected with the excitation laser at 514.5 nm (CW Ar/Kr ion laser), substantiate the possibility of evaluating the local temperature, which has been tested in the range of 298 – 313 K, a range of interest for biological applications. The power of the laser has been carefully chosen in order to avoid eventual heating due to the laser irradiation. The data show that TiO2 NPs possess a high sensitivity and low uncertainty in the range of a few degrees as Raman nanothermometer material.

Autonomous navigation in underwater environments presents challenges due to factors such as light absorption and water turbidity, limiting the effectiveness of optical sensors. Sonar systems are commonly used for perception in underwater operations as they are unaffected by these limitations. Traditional computer vision algorithms are less effective when applied to sonar-generated acoustic images, while convolutional neural networks (CNNs) typically require large amounts of labeled training data that are often unavailable or difficult to acquire. To this end, we propose a novel compact deep sonar descriptor pipeline that can generalize to real scenarios while being trained exclusively on synthetic data. Our architecture is based on a ResNet18 back-end and a properly parameterized random Gaussian projection layer, whereas input sonar data is enhanced with standard ad-hoc normalization/prefiltering techniques. A customized synthetic data generation procedure is also presented. The proposed method has been evaluated extensively using both synthetic and publicly available real data, demonstrating its effectiveness compared to state-of-the-art methods.

We study the complexity of local graph-centrality estimation, with the goal of approximating the centrality score of a given target node while exploring only a sublinear number of nodes/arcs of the graph and performing a sublinear number of elementary operations. We develop a technique, which we apply to PageRank and Heat Kernel, for constructing a low-variance score estimator through a local exploration of the graph. We obtain an algorithm that, given any node in any graph of n nodes and m arcs, with probability (1-delta) computes a multiplicative (1pmepsilon)-approximation of its score by examining only O~(min(n1/2Delta1/2,n1/2m1/4)) nodes/arcs, where Delta is the maximum outdegree of the graph and poly(epsilon-1) and polylog(delta-1) factors are omitted for readability. A similar bound holds for computational cost. We also prove a lower bound of Omega(min(n1/2Delta1/2, n1/3m1/3)) for both query complexity and computational complexity. Moreover, in the jump-and-crawl graph-access model, our technique yields a O~(min(n1/2Delta1/2,n2/3))-queries algorithm; we show that this algorithm is optimal up to a logarithmic factor-in fact, sublogarithmic in the case of PageRank. These are the first algorithms with sublinear worst-case bounds for general directed graphs and any choice of the target node.

Due to the increasing number of people with severe obesity, the demand for patient-specific modelling in bariatric surgery (BS) is increasing because its potentialities in the improvement of surgical planning, optimization of outcomes and prediction of the mechanical response of the stomach. However, the patient-specific anatomical reconstruction is a pivotal and often time-consuming step due to the lack of efficient and fully automatized tools. Ongoing studies on multi-organ segmentation methods based on neural networks for magnetic resonance images (MRI) are currently available, but they have still several limits, mainly due to both the highly flexible individuals anatomical properties, and convolutional neural networks (CNN) trained only in the detection of physiological stomachs. The aim of this work is to perform a convenient transfer learning from a general-purpose CNN, able to improve the performance in automatically detecting the stomach region of patients with severe obesity. The proposed approach represents the basis for the development of pre- and post-surgical computational models for rapid clinical analysis, especially to boost the mechanical stimulation of gastric receptors. The segmentation masks and the corresponding 3D models were compared with the corresponding manual MRI segmentation as ground truth. Intersection Over Union (IOU) and DICE coefficients (DICE) were used to evaluate 2D masks segmentation, while the Relative Volume Error (RVE), mean surface distance (MSD), standard deviation and the Normalised Hausdorff distance (NHD) were applied to assess the obtained 3D results.

Next-generation telescopes, like Euclid, Rubin/LSST, and Roman, will open new windows on the Universe, allowing us to infer physical properties for tens of millions of galaxies. Machine-learning methods are increasingly becoming the most efficient tools to handle this enormous amount of data, because they are often faster and more accurate than traditional methods. We investigate how well redshifts, stellar masses, and star-formation rates (SFRs) can be measured with deep-learning algorithms for observed galaxies within data mimicking the Euclid and Rubin/LSST surveys. We find that deep-learning neural networks and convolutional neural networks (CNNs), which are dependent on the parameter space of the training sample, perform well in measuring the properties of these galaxies and have a better accuracy than methods based on spectral energy distribution fitting. CNNs allow the processing of multiband magnitudes together with HE-band images. We find that the estimates of stellar masses improve with the use of an image, but those of redshift and SFR do not. Our best results are deriving (i) the redshift within a normalized error of 3 in the HE band; (ii) the stellar mass within a factor of two (∼0.3 dex) for 99.5 per cent of the considered galaxies; and (iii) the SFR within a factor of two (∼0.3 dex) for ∼70 per cent of the sample. We discuss the implications of our work for application to surveys as well as how measurements of these galaxy parameters can be improved with deep learning.