• Rural Digital Europe
• Publications close
• HAL-Inserm close

Automatic surgical instrument segmentation of endoscopic images is a crucial building block of many computer-assistance applications for minimally invasive surgery. So far, state-of-the-art approaches completely rely on the availability of a ground-truth supervision signal, obtained via manual annotation, thus expensive to collect at large scale. In this paper, we present FUN-SIS, a Fully-UNsupervised approach for binary Surgical Instrument Segmentation. FUN-SIS trains a per-frame segmentation model on completely unlabelled endoscopic videos, by solely relying on implicit motion information and instrument shape-priors. We define shape-priors as realistic segmentation masks of the instruments, not necessarily coming from the same dataset/domain as the videos. The shape-priors can be collected in various and convenient ways, such as recycling existing annotations from other datasets. We leverage them as part of a novel generative-adversarial approach, allowing to perform unsupervised instrument segmentation of optical-flow images during training. We then use the obtained instrument masks as pseudo-labels in order to train a per-frame segmentation model; to this aim, we develop a learning-from-noisy-labels architecture, designed to extract a clean supervision signal from these pseudo-labels, leveraging their peculiar noise properties. We validate the proposed contributions on three surgical datasets, including the MICCAI 2017 EndoVis Robotic Instrument Segmentation Challenge dataset. The obtained fully-unsupervised results for surgical instrument segmentation are almost on par with the ones of fully-supervised state-of-the-art approaches. This suggests the tremendous potential of the proposed method to leverage the great amount of unlabelled data produced in the context of minimally invasive surgery.

International audience; Owing to the complexity of obtaining the landslide inventory data, it is a challenge to establish a landslide spatial prediction model with limited labeled samples. This paper proposed a novel strategy, namely transfer learning with attributes (TLAs), to make good use of existing landslide inventory data, a strategy that is based on a variational autoencoder of a generative adversarial network (VAEGAN) for improving the landslide spatial prediction performance in sample-scarce areas. Different from transfer learning (TL), TLAs are pretraining the model with the data reconstructed by VAEGAN, so that the models learn in advance the landslide attributes of sample-scarce areas. Accordingly, a database containing a total of 986 landslides in three study areas with 14 landslide-influencing factors was established, and each of the three models, i.e., convolutional neural networks (CNNs), bidirectional long short-term memory (BiLSTM) and gated recurrent units (GRUs), was respectively selected as the feature extractor of the VAEGAN to reconstruct the data with attributes and the prediction model to generate the landslide susceptibility maps to investigate and validate the proposed TLA strategy. The experimental results showed that the TLA strategy increased the mean value of evaluators, such as the area under the receiver-operating characteristic (AUROC), F1-score, precision, recall and accuracy by about 2–7% compared with TL, results that indicated that the generated data have the attribute of specific study areas and the effectiveness of TLA strategy in sample-scare areas.

International audience; Because our civilization has relied on pesticides to fight weeds, insects, and diseases since antiquity, the use of these chemicals has become natural and exclusive. Unfortunately, the use of pesticides has progressively had alarming effects on water quality, biodiversity, and humanhealth. This paper proposes to improve farming practices by replacing pesticides with a laser-based robotic approach. This study focused on the neutralization of aphids, as they are among the most harmful pests for crops and complex to control. With the help of deep learning, we developeda mobile robot that spans crop rows, locates aphids, and neutralizes them with laser beams. We have built a prototype with the sole purpose of validating the localization-neutralization loop on a single seedling row. The experiments performed in our laboratory demonstrate the feasibility ofdetecting different lines of aphids (50% detected at 3 cm/s) and of neutralizing them (90% mortality) without impacting the growth of their host plants. The results are encouraging since aphids are one of the most challenging crop pests to eradicate. However, enhancements in detection and mainly intargeting are necessary to be useful in a real farming context. Moreover, robustness regarding field conditions should be evaluated.

ObjectivesGiven mixed evidence for carcinogenicity of current-use herbicides, we studied the relationship between occupational herbicide use and risk of non-Hodgkin’s lymphoma (NHL) in a large, pooled study.MethodsWe pooled data from 10 case-control studies participating in the International Lymphoma Epidemiology Consortium, including 9229 cases and 9626 controls from North America, the European Union and Australia. Herbicide use was coded from self-report or by expert assessment in the individual studies, for herbicide groups (eg, phenoxy herbicides) and active ingredients (eg, 2,4-dichlorophenoxyacetic acid (2,4-D), glyphosate). The association between each herbicide and NHL risk was estimated using logistic regression to produce ORs and 95% CIs, with adjustment for sociodemographic factors, farming and other pesticides.ResultsWe found no substantial association of all NHL risk with ever-use of any herbicide (OR=1.10, 95% CI: 0.94 to 1.29), nor with herbicide groups or active ingredients. Elevations in risk were observed for NHL subtypes with longer duration of phenoxy herbicide use, such as for any phenoxy herbicide with multiple myeloma (>25.5 years, OR=1.78, 95% CI: 0.74 to 4.27), 2,4-D with diffuse large B-cell lymphoma (>25.5 years, OR=1.47, 95% CI: 0.67 to 3.21) and other (non-2,4-D) phenoxy herbicides with T-cell lymphoma (>6 years, lagged 10 years, OR=3.24, 95% CI: 1.03 to 10.2). An association between glyphosate and follicular lymphoma (lagged 10 years: OR=1.48, 95% CI: 0.98 to 2.25) was fairly consistent across analyses.ConclusionsMost of the herbicides examined were not associated with NHL risk. However, associations of phenoxy herbicides and glyphosate with particular NHL subtypes underscore the importance of estimating subtype-specific risks.

International audience; Clustering is an essential tool for data analysis and visualization. It is particularly useful in case of a lack of labels, which prevent the use of supervised methods. The analysis of satellite images is particularly prone to this problem, especially when studied as time series, because the access to this type of data is still recent. Among all clustering methods, the ones based on Deep Neural Networks (DNNs) have seen an increasing interest lately, but only a few works have been conducted on time series yet. This paper aims to give more insight on how current clustering methods based on DNNs can be applied to Satellite Images Time Series (SITS) and it shows that with a proper configuration they can perform better compared to classical non-deep methods.

AbstractThe etiology of central nervous system (CNS) tumors is complex and involves many suspected risk factors. Scientific evidence remains insufficient, in particular in the agricultural field. The goal of our study was to investigate associations between agricultural activities and CNS tumors in the entire French farm manager workforce using data from the TRACTOR project. The TRACTOR project hold a large administrative health database covering the entire French agricultural workforce, over the period 2002‐2016, on the whole French metropolitan territory. Associations were estimated for 26 activities and CNS tumors using Cox proportional hazards model, with time to first CNS tumor insurance declaration as the underlying timescale, adjusting for sex, age and geographical area. There were 1017 cases among 1 036 069 farm managers, including 317 meningiomas and 479 gliomas. Associations varied with tumor types, sex and types of crop and animal farming. Analyses showed several increased risks of CNS tumors, in particular for animal farming. The main increases in risk were observed for meningioma in mixed dairy and cow farming (hazard ratio [HR] = 1.75, 95% confidence interval [CI]: 1.09‐2.81) and glioma in pig farming (HR = 2.28, 95% CI: 1.37‐3.80). Our study brings new insights on the association of a wide range of agricultural activities and CNS tumor and subtype‐specific risks in farm managers. Although these findings need to be corroborated in further studies and should be interpreted cautiously, they could have implications for enhancing CNS tumor surveillance in agriculture.

International audience

AbstractBackgroundFor many co‐manipulative applications, variable damping is a valuable feature provided by robots. One approach is implementing a high viscosity at low velocities and a low viscosity at high velocities. This, however, is proven to have the possibility to alter human natural motion performance.MethodsWe show that the distortion is caused by the viscosity drop resulting in robot's resistance to motion. To address this, a method for stably achieving the desired behaviour is presented. It involves leveraging a first‐order linear filter to slow the viscosity variation down.ResultsThe proposition is supported by a theoretical analysis using a robotic model. Meanwhile, the user performance in human‐robot experiments gets significantly improved, showing the practical efficiency in real applications.ConclusionsThis paper discusses the variable viscosity control in the context of co‐manipulation. An instability problem and its solution were theoretically shown and experimentally evidenced through human‐robot experiments.