Dissecting self-supervised learning methods for surgical computer vision.

The field of surgical computer vision has undergone considerable breakthroughs in recent years with the rising popularity of deep neural network-based methods. However, standard fully-supervised approaches for training such models require vast amounts of annotated data, imposing a prohibitively high cost; especially in the clinical domain. Self-Supervised Learning (SSL) methods, which have begun to gain traction in the general computer vision community, represent a potential solution to these annotation costs, allowing to learn useful representations from only unlabeled data. Still, the effectiveness of SSL methods in more complex and impactful domains, such as medicine and surgery, remains limited and unexplored. In this work, we address this critical need by investigating four state-of-the-art SSL methods (MoCo v2, SimCLR, DINO, SwAV) in the context of surgical computer vision. We present an extensive analysis of the performance of these methods on the Cholec80 dataset for two fundamental and popular tasks in surgical context understanding, phase recognition and tool presence detection. We examine their parameterization, then their behavior with respect to training data quantities in semi-supervised settings. Correct transfer of these methods to surgery, as described and conducted in this work, leads to substantial performance gains over generic uses of SSL - up to 7.4% on phase recognition and 20% on tool presence detection - as well as state-of-the-art semi-supervised phase recognition approaches by up to 14%. Further results obtained on a highly diverse selection of surgical datasets exhibit strong generalization properties. The code is available at https://github.com/CAMMA-public/SelfSupSurg.

Prediction of cesarean delivery in class III obese nulliparous women: An externally validated model using machine learning.

BACKGROUND: class III obese women, are at a higher risk of cesarean section during labor, and cesarean section is responsible for increased maternal and neonatal morbidity in this population. OBJECTIVE: the objective of this project was to develop a method with which to quantify cesarean section risk before labor. METHODS: this is a multicentric retrospective cohort study conducted on 410 nulliparous class III obese pregnant women who attempted vaginal delivery in two French university hospitals. We developed two predictive algorithms (a logistic regression and a random forest models) and assessed performance levels and compared them. RESULTS: the logistic regression model found that only initial weight and labor induction were significant in the prediction of unplanned cesarean section. The probability forest was able to predict cesarean section probability using only two pre-labor characteristics: initial weight and labor induction. Its performances were higher and were calculated for a cut-point of 49.5% risk and the results were (with 95% confidence intervals): area under the curve 0.70 (0.62,0.78), accuracy 0.66 (0.58, 0.73), specificity 0.87 (0.77, 0.93), and sensitivity 0.44 (0.32, 0.55). CONCLUSIONS: this is an innovative and effective approach to predicting unplanned CS risk in this population and could play a role in the choice of a trial of labor versus planned cesarean section. Further studies are needed, especially a prospective clinical trial. FUNDING: French state funds "Plan Investissements d'Avenir" and Agence Nationale de la Recherche.

Predicting COVID-19 positivity and hospitalization with multi-scale graph neural networks.

The pandemic of COVID-19 is undoubtedly one of the biggest challenges for modern healthcare. In order to analyze the spatio-temporal aspects of the spread of COVID-19, technology has helped us to track, identify and store information regarding positivity and hospitalization, across different levels of municipal entities. In this work, we present a method for predicting the number of positive and hospitalized cases via a novel multi-scale graph neural network, integrating information from fine-scale geographical zones of a few thousand inhabitants. By leveraging population mobility data and other features, the model utilizes message passing to model interaction between areas. Our proposed model manages to outperform baselines and deep learning models, presenting low errors in both prediction tasks. We specifically point out the importance of our contribution in predicting hospitalization since hospitals became critical infrastructure during the pandemic. To the best of our knowledge, this is the first work to exploit high-resolution spatio-temporal data in a multi-scale manner, incorporating additional knowledge, such as vaccination rates and population mobility data. We believe that our method may improve future estimations of positivity and hospitalization, which is crucial for healthcare planning.

Solid harmonic wavelet scattering for predictions of molecule properties.

We present a machine learning algorithm for the prediction of molecule properties inspired by ideas from density functional theory (DFT). Using Gaussian-type orbital functions, we create surrogate electronic densities of the molecule from which we compute invariant "solid harmonic scattering coefficients" that account for different types of interactions at different scales. Multilinear regressions of various physical properties of molecules are computed from these invariant coefficients. Numerical experiments show that these regressions have near state-of-the-art performance, even with relatively few training examples. Predictions over small sets of scattering coefficients can reach a DFT precision while being interpretable.

Discrete Sparse Coding.

Sparse coding algorithms with continuous latent variables have been the subject of a large number of studies. However, discrete latent spaces for sparse coding have been largely ignored. In this work, we study sparse coding with latents described by discrete instead of continuous prior distributions. We consider the general case in which the latents (while being sparse) can take on any value of a finite set of possible values and in which we learn the prior probability of any value from data. This approach can be applied to any data generated by discrete causes, and it can be applied as an approximation of continuous causes. As the prior probabilities are learned, the approach then allows for estimating the prior shape without assuming specific functional forms. To efficiently train the parameters of our probabilistic generative model, we apply a truncated expectation-maximization approach (expectation truncation) that we modify to work with a general discrete prior. We evaluate the performance of the algorithm by applying it to a variety of tasks: (1) we use artificial data to verify that the algorithm can recover the generating parameters from a random initialization, (2) use image patches of natural images and discuss the role of the prior for the extraction of image components, (3) use extracellular recordings of neurons to present a novel method of analysis for spiking neurons that includes an intuitive discretization strategy, and (4) apply the algorithm on the task of encoding audio waveforms of human speech. The diverse set of numerical experiments presented in this letter suggests that discrete sparse coding algorithms can scale efficiently to work with realistic data sets and provide novel statistical quantities to describe the structure of the data.