Search

Background Whether Peroral Endoscopic Myotomy (POEM) can be proposed as a second-line treatment in patients with achalasia remains to be confirmed in real-life series. Objective This study aimed to compare the efficacy, feasibility and safety of POEM between treatment-naïve patients and patients who had prior endoscopic or surgical therapies for achalasia. Methods All consecutive patients who underwent a POEM procedure for achalasia in our centre from June 2015 to September 2018 were included in this retrospective study. They were classified into treatment-naïve patients (POEM1) and patients who had at least one previous endoscopic and/or surgical treatment for achalasia (POEM2). Results A total of 105 patients were included, 52 in the POEM1 group and 53 in the POEM2 group. Clinical success (defined as an Eckardt score ≤ 3) at 6 months was observed in 93% of POEM1 patients and 84% of POEM2 patients (p = 0.18). Technical success rate was not significantly different between the two groups (100% vs 96%, respectively; p = 0.50). No significant difference was noted in terms of adverse event rate (19% vs 19%, respectively; p = 1.00). Post-procedure pain occurred in 12% of treatment-naive and 9% of non-naïve patients (p = 0.76). The median length of hospital stay was 3 days in both groups (p = 0.17). Symptomatic gastroesophageal reflux occurred in 25% of POEM1 patients and 16% of POEM2 patients (p = 0.24). Conclusion Efficacy, feasibility and safety of POEM are not different between treatment-naïve and non-naïve patients. POEM is a valuable second-line approach in patients with persistent symptoms of achalasia after surgical or endoscopic treatments.

Data on outcomes of microsatellite instable and/or mismatch repair deficient (dMMR/MSI) digestive non-colorectal tumors are limited.To evaluate overall survival (OS) of patients with dMMR/MSI digestive non-colorectal tumor.All consecutive patients with a dMMR/MSI digestive non-colorectal tumor were included in this French retrospective multicenter study.One hundred and sixteen patients were included with a mean age of 63.6 years and 32.6% with a Lynch syndrome. Most tumors were oesophago-gastric (54.3%) or small bowel (32.8%) adenocarcinomas and at a localized stage at diagnosis (86.7%). In patients with localized tumors and R0 resection, median OS was 134.0 ± 64.2 months. Median disease-free survival (DFS) was 100.3 ± 65.7 months. Considering oesophago-gastric tumors, median DFS was improved when chemotherapy was added to surgery (not reached versus 22.8 ± 10.0 months, p = 0.03). In patients with advanced tumors treated by chemotherapy, median OS was 14.2 ± 1.9 months and median progression-free survival was 7.4 ± 1.6 months.dMMR/MSI digestive non-colorectal tumors are mostly diagnosed at a non-metastatic stage with a good prognosis. Advanced dMMR/MSI digestive non-colorectal tumors have a poor prognosis with standard chemotherapy.

Endoscopic management is preferred to surgical management for large superficial colorectal lesions. However, the optimal endoscopic resection strategy (piecemeal endoscopic mucosal resection [pEMR] or endoscopic submucosal dissection [ESD]) is still debated from an economical point of view. To date, in France, there is no Health Insurance reimbursement rate for the hospital stays related to ESD. We searched to estimate the global cost of colorectal ESD and to define the most cost-effectiveness endoscopic strategy.A model was created to compare the cost-effectiveness of ESD and pEMR according to optical diagnosis (Japan NBI Expert Team [JNET], laterally spreading tumour [LST], CONECCT). We distinguished three groups from the same multicentre ESD cohort and compared the medical and economic outcomes: real-life ESD data (Universal-ESD or U-ESD) compared to modelled selective ESD (S-ESD JNET; S-ESD LST; S-ESD CONECCT) and exclusive pEMR strategies (Universal-EMR or U-EMR).The en-bloc, R0, and curative resection rates were 97.5%, 86.5%, and 82.6%, respectively in the real life French ESD cohort of 833 colorectal lesions. U-ESD was the least-expensive strategy, with a global cost of 2,858,048.17 €, i.e. 3,431.03 €/patient and was also the most effective strategy because it avoided 774 surgeries, which was more than any other strategy. It outperformed S-ESD CONNECT (global cost = 2,951,411.44 €, and 3,543.11 €/patient, 765 surgeries avoided, S-ESD LST (global cost = 3,055,951.53 €, and 3,668.61 €/patient, 749 surgeries avoided), and S-ESD JNET (global cost = 3,547,426.97 € and 4,258.62 €/patient, 704 surgeries avoided) and U-EMR (global cost = 4,060,547.62 € and 4,874.61 €/patient, 620 surgeries avoided). Even though a model which optimized pEMR results (0% technical failure, 0% primary surgery), U-EMR strategy remained the most expansive strategy and the one that avoided the least surgeries.ESD for all LSTs upper than 20 mm is more cost-effective than pEMR, and S-ESD.

Simple Summary Lynch syndrome is characterized by a higher relative risk of developing certain cancers, especially digestive cancers. Many guidelines from different scientific societies are now available and allow excellent follow-up for these patients, but occasionally propose divergent management approaches. We provide here a synthesis of these guidelines and a focus on prevention, diagnosis, and endoscopic follow-up of these digestive cancers and early neoplasia. Abstract Lynch syndrome patients could benefit from various recommendations to prevent digestive cancers. In this review, we summarize the criteria to identify Lynch syndrome in patients with digestive cancers. We detail endoscopic screening procedures in patients with Lynch syndrome for gastric, small bowel, pancreatic, and colorectal cancers. We review the precise modalities of endoscopic follow-up, particularly the discrepancies that exist between the guidelines of the various scientific societies. We discuss the treatment of colorectal cancers in Lynch syndrome cases and patient adherence to endoscopic follow-up programs.

Recent work suggests that adversarial examples are enabled by high-frequency components in the dataset. In the speech domain where spectrograms are used extensively, masking those components seems like a sound direction for defenses against attacks. We explore a smoothing approach based on additive noise masking in priority high frequencies. We show that this approach is much more robust than the naive noise filtering approach, and a promising research direction. We successfully apply our defense on a Librispeech speaker identification task, and on the UrbanSound8K audio classification dataset.

Many machine learning tasks can be posed as matching problems in which we are given a “probe” entry that we expect matches some of the entries in our “gallery”. The general solution to these problems is to retrieve matching entries based on statistical dependencies between the probe and the gallery data that are learned using complex models. Often, however, there are other common covariates to the probe and gallery data which might be easily inferred and may explain some of the statistical dependencies between the two. In this paper we present a probabilistic framework to derive optimal matching strategies based only on covariate features for three broad tasks, namely N-way classification, pairwise verification and ranking. We use canonical metrics to determine the maximum performance that can be expected if only covariate features are used and determine the marginal gain of using complex models. We find that covariate matching achieves an EER within 10% of a CNN in the verification task, and an MAP within 22% of the a DNN based model in the ranking task.

International audience; Full-field optical coherence tomography (FFOCT) is an imaging technique of biological tissue based on tissue light reflectance analysis. We evaluated the feasibility of imaging fresh digestive mucosal biopsies after a quick mounting procedure (5 min) using two distinct modalities of FFOCT. In static FFOCT mode, we gained high-resolution images of general gut tissue-specific architecture, such as oesophageal papillae, gastric pits, duodenal villi and colonic crypts. In dynamic FFOCT mode, we imaged individual epithelial cells of the mucosal lining with a cellular or subcellular resolution and identified cellular components of the lamina propria. FFOCT represents a promising dye-free imaging tool for on-site analysis of gut tissue remodelling.

Deploying deep learning models, comprising of non-linear combination of millions, even billions, of parameters is challenging given the memory, power and compute constraints of the real world. This situation has led to research into model compression techniques most of which rely on suboptimal heuristics and do not consider the parameter redundancies due to linear dependence between neuron activations in overparametrized networks. In this paper, we propose a novel model compression approach based on exploitation of linear dependence, that compresses networks by elimination of entire neurons and redistribution of their activations over other neurons in a manner that is provably lossless while training. We combine this approach with an annealing algorithm that may be applied during training, or even on a trained model, and demonstrate, using popular datasets, that our method results in a reduction of up to 99\% in overall network size with small loss in performance. Furthermore, we provide theoretical results showing that in overparametrized, locally linear (ReLU) neural networks where redundant features exist, and with correct hyperparameter selection, our method is indeed able to capture and suppress those dependencies.

International audience; Background: Analysis of serum biomarkers for the assessment of atrophic gastritis (AG), considered as gastric precancerous lesion, is of growing interest and recommended by current guidelines. Our aim was to evaluate the diagnostic performance of a panel of biomarkers (GastroPanel®) for the detection of AG in France, a country of a low gastric cancer (GC) incidence. Material and Methods: In this prospective, multicenter, cross-sectional study, consecutive patients considered at increased risk of GC and undergoing upper endoscopy with gastric biopsies were included. Blood samples were collected for the analysis of GastroPanel® (association of Pepsinogens I and II, Gastrin-17, and Helicobacter pylori serology) using ELISA. The results of GastroPanel® were compared to the results of histology considered as the reference. Results: Between 2016 and 2019, 344 patients (148 cases with AG, 196 controls without AG) were included. Sensitivity, specificity, positive, and negative predictive values for the detection of AG by GastroPanel® were of 39.9% (95% CI 31.9; 48.2), 93.4% (95% CI 88.9; 96.4), 81.9 (95% CI 71.1; 90.0), and 67.3 (95% CI 61.4; 72.8), respectively. The sensitivity was significantly higher for the detection of severe AG [60.8% (95% CI 46.1; 74.6) P =.015] and corpus AG [61.0% (95% CI 49.2; 72.0), P =.004]. Diagnostic performances of GastroPanel® tended to be better than those of Pepsinogen I alone, but the difference did not reach statistical significance (P =.068). Conclusion: Serum pepsinogen and GastroPanel® tests show promising results for the detection of AG, especially of corpus AG and severe AG, in patients at high risk of GC in France.

Using transfer learning to help in solving a new classification task where labeled data is scarce is becoming popular. Numerous experiments with deep neural networks, where the representation learned on a source task is transferred to learn a target neural network, have shown the benefits of the approach. This paper, similarly, deals with hypothesis transfer learning. However, it presents a new approach where, instead of transferring a representation, the source hypothesis is kept and this is a translation from the target domain to the source domain that is learned. In a way, a change of representation is learned. We show how this method performs very well on a classification of time series task where the space of time series is changed between source and target.

International audience

International audience; Aims - Spina Bifida (SB) is a rare congenital condition that frequently impairs the neurological control of both fecal continence and defecation. Several therapeutic strategies have been proposed but impact assessment is lacking. Our objectives were to quantify the symptomatic improvement and to determine the optimal strategy in this rare condition where randomized controlled trials are difficult to conduct. Methods - Data were extracted from a prospective database. The present analysis focused on patients having undergone at least two gastroenterological assessments. A standardized therapeutic approach was used from the first visit. Improvement was quantified by the variation of quantified symptomatic scores. Results - The data of of 57 adults with SB (gender F/M: 30/27 [52.6/47.4%]; mean age: 33.8 [18.5] years) were extracted. After a mean follow-up of 46 months, 23/57 patients (40.4%) had at least improvement of one point of the Cleveland Clinic Incontinence score (CCIS); 13/57 (22.8%) reported a significant improvement of continence (delta score >50%). Five of the twelve patients (41.6%) with CCIS

In models to generate program source code from natural language, representing this code in a tree structure has been a common approach. However, existing methods often fail to generate complex code correctly due to a lack of ability to memorize large and complex structures. We introduce ReCode, a method based on subtree retrieval that makes it possible to explicitly reference existing code examples within a neural code generation model. First, we retrieve sentences that are similar to input sentences using a dynamic-programming-based sentence similarity scoring method. Next, we extract n-grams of action sequences that build the associated abstract syntax tree. Finally, we increase the probability of actions that cause the retrieved n-gram action subtree to be in the predicted code. We show that our approach improves the performance on two code generation tasks by up to +2.6 BLEU., Comment: This paper is accepted in EMNLP 2018. It has 6 pages

Summary form only given. In the last few years the impact of mobile communications in railway environment has increased, showing the vital importance of these systems for management and control of train traffic and also for the communications services of the passengers. Therefore, this work has been focused on the study of propagation in a key element of rail transport -a tunnel. The propagation on radio frequencies in tunnels is complex and depends of several parameters therefore is very important to get an accurate modeling to understand, improve and thus stimulate the implementation of new communications systems. Radio coverage of tunnels using antennas is much more flexible and less expensive than the use of leaky feeder but requires an accurate modeling of the propagation and this modeling has to be validated using measurements. Consequently, comprehensive measurement campaign has been carried out in various types of tunnels of Metro Madrid to achieve this goal. Specifically, different segments such as straight curve and passing through station in metropolitan lines has been measured. These measurements were carried out in three different frequency bands: 900 MHz, 2.4GHz and 5.8 GHz commonly used by mobile communication. The results have been processed extracting principal statistics.