Annotation-free learning of a spatio-temporal manifold of the cell life cycle

The cell cycle is a complex biological phenomenon, which plays an important role in many cell biological processes and disease states. Machine learning is emerging to be a pivotal technique for the study of the cell cycle, resulting in a number of available tools and models for the analysis of the cell cycle. Most, however, heavily rely on expert annotations, prior knowledge of mechanisms, and imaging with several fluorescent markers to train their models. Many are also limited to processing only the spatial information in the cell images. In this work, we describe a different approach based on representation learning to construct a manifold of the cell life cycle. We trained our model such that the representations are learned without exhaustive annotations nor assumptions. Moreover, our model uses microscopy images derived from a single fluorescence channel and utilizes both the spatial and temporal information in these images. We show that even with fewer channels and self-supervision, information relevant to cell cycle analysis such as staging and estimation of cycle duration can still be extracted, which demonstrates the potential of our approach to aid future cell cycle studies and in discovery cell biology to probe and understand novel dynamic systems.

Self-Supervision for Medical Image Classification: State-of-the-Art Performance with ~100 Labeled Training Samples per Class

Is self-supervised deep learning (DL) for medical image analysis already a serious alternative to the de facto standard of end-to-end trained supervised DL? We tackle this question for medical image classification, with a particular focus on one of the currently most limiting factor of the field: the (non-)availability of labeled data. Based on three common medical imaging modalities (bone marrow microscopy, gastrointestinal endoscopy, dermoscopy) and publicly available data sets, we analyze the performance of self-supervised DL within the self-distillation with no labels (DINO) framework. After learning an image representation without use of image labels, conventional machine learning classifiers are applied. The classifiers are fit using a systematically varied number of labeled data (1-1000 samples per class). Exploiting the learned image representation, we achieve state-of-the-art classification performance for all three imaging modalities and data sets with only a fraction of between 1% and 10% of the available labeled data and about 100 labeled samples per class.

Self-supervised denoising of Nyquist-sampled volumetric images via deep learning

Purpose

Deep learning has demonstrated excellent performance enhancing noisy or degraded biomedical images. However, many of these models require access to a noise-free version of the images to provide supervision during training, which limits their utility. Here, we develop an algorithm (noise2Nyquist) that leverages the fact that Nyquist sampling provides guarantees about the maximum difference between adjacent slices in a volumetric image, which allows denoising to be performed without access to clean images. We aim to show that our method is more broadly applicable and more effective than other self-supervised denoising algorithms on real biomedical images, and provides comparable performance to algorithms that need clean images during training.

Approach

We first provide a theoretical analysis of noise2Nyquist and an upper bound for denoising error based on sampling rate. We go on to demonstrate its effectiveness in denoising in a simulated example as well as real fluorescence confocal microscopy, computed tomography, and optical coherence tomography images.

Results

We find that our method has better denoising performance than existing self-supervised methods and is applicable to datasets where clean versions are not available. Our method resulted in peak signal to noise ratio (PSNR) within 1 dB and structural similarity (SSIM) index within 0.02 of supervised methods. On medical images, it outperforms existing self-supervised methods by an average of 3 dB in PSNR and 0.1 in SSIM.

Conclusion

noise2Nyquist can be used to denoise any volumetric dataset sampled at at least the Nyquist rate making it useful for a wide variety of existing datasets.

Surgical Gesture Recognition in Laparoscopic Tasks Based on the Transformer Network and Self-Supervised Learning

In this study, we propose a deep learning framework and a self-supervision scheme for video-based surgical gesture recognition. The proposed framework is modular. First, a 3D convolutional network extracts feature vectors from video clips for encoding spatial and short-term temporal features. Second, the feature vectors are fed into a transformer network for capturing long-term temporal dependencies. Two main models are proposed, based on the backbone framework: C3DTrans (supervised) and SSC3DTrans (self-supervised). The dataset consisted of 80 videos from two basic laparoscopic tasks: peg transfer (PT) and knot tying (KT). To examine the potential of self-supervision, the models were trained on 60% and 100% of the annotated dataset. In addition, the best-performing model was evaluated on the JIGSAWS robotic surgery dataset. The best model (C3DTrans) achieves an accuracy of 88.0%, a 95.2% clip level, and 97.5% and 97.9% (gesture level), for PT and KT, respectively. The SSC3DTrans performed similar to C3DTrans when training on 60% of the annotated dataset (about 84% and 93% clip-level accuracies for PT and KT, respectively). The performance of C3DTrans on JIGSAWS was close to 76% accuracy, which was similar to or higher than prior techniques based on a single video stream, no additional video training, and online processing.

Handcrafted versus non-handcrafted (self-supervised) features for the classification of antimicrobial peptides: complementary or redundant?

Antimicrobial peptides (AMPs) have received a great deal of attention given their potential to become a plausible option to fight multi-drug resistant bacteria as well as other pathogens. Quantitative sequence-activity models (QSAMs) have been helpful to discover new AMPs because they allow to explore a large universe of peptide sequences and help reduce the number of wet lab experiments. A main aspect in the building of QSAMs based on shallow learning is to determine an optimal set of protein descriptors (features) required to discriminate between sequences with different antimicrobial activities. These features are generally handcrafted from peptide sequence datasets that are labeled with specific antimicrobial activities. However, recent developments have shown that unsupervised approaches can be used to determine features that outperform human-engineered (handcrafted) features. Thus, knowing which of these two approaches contribute to a better classification of AMPs, it is a fundamental question in order to design more accurate models. Here, we present a systematic and rigorous study to compare both types of features. Experimental outcomes show that non-handcrafted features lead to achieve better performances than handcrafted features. However, the experiments also prove that an improvement in performance is achieved when both types of features are merged. A relevance analysis reveals that non-handcrafted features have higher information content than handcrafted features, while an interaction-based importance analysis reveals that handcrafted features are more important. These findings suggest that there is complementarity between both types of features. Comparisons regarding state-of-the-art deep models show that shallow models yield better performances both when fed with non-handcrafted features alone and when fed with non-handcrafted and handcrafted features together.

Modeling Cognitive Load as a Self-Supervised Brain Rate with Electroencephalography and Deep Learning

The principal reason for measuring mental workload is to quantify the cognitive cost of performing tasks to predict human performance. Unfortunately, a method for assessing mental workload that has general applicability does not exist yet. This is due to the abundance of intuitions and several operational definitions from various fields that disagree about the sources or workload, its attributes, the mechanisms to aggregate these into a general model and their impact on human performance. This research built upon these issues and presents a novel method for mental workload modelling from EEG data employing deep learning. This method is self-supervised, employing a continuous brain rate, an index of cognitive activation, and does not require human declarative knowledge. The aim is to induce models automatically from data, supporting replicability, generalisability and applicability across fields and contexts. This specific method is a convolutional recurrent neural network trainable with spatially preserving spectral topographic head-maps from EEG data, aimed at fitting a novel brain rate variable. Findings demonstrate the capacity of the convolutional layers to learn meaningful high-level representations from EEG data since within-subject models had, on average, a test Mean Absolute Percentage Error of around 11%. The addition of a Long-Short Term Memory layer for handling sequences of high-level representations was not significant, although it did improve their accuracy. These findings point to the existence of quasi-stable blocks of automatically learnt high-level representations of cognitive activation because they can be induced through convolution and seem not to be dependent on each other over time, intuitively matching the non-stationary nature of brain responses. Additionally, across-subject models, induced with data from an increasing number of participants, thus trained with data containing more variability, obtained a similar accuracy to the within-subject models. This highlights the potential generalisability of the induced high-level representations across people, suggesting the existence of subject-independent cognitive activation patterns. This research contributes to the body of knowledge by providing scholars with a novel computational method for mental workload modelling that aims to be generally applicable and does not rely on ad hoc human crafted models.

Noise2Kernel: Adaptive Self-Supervised Blind Denoising Using a Dilated Convolutional Kernel Architecture

With the advent of unsupervised learning, efficient training of a deep network for image denoising without pairs of noisy and clean images has become feasible. Most current unsupervised denoising methods are built on self-supervised loss with the assumption of zero-mean noise under the signal-independent condition, which causes brightness-shifting artifacts on unconventional noise statistics (i.e., different from commonly used noise models). Moreover, most blind denoising methods require a random masking scheme for training to ensure the invariance of the denoising process. In this study, we propose a dilated convolutional network that satisfies an invariant property, allowing efficient kernel-based training without random masking. We also propose an adaptive self-supervision loss to increase the tolerance for unconventional noise, which is specifically effective in removing salt-and-pepper or hybrid noise where prior knowledge of noise statistics is not readily available. We demonstrate the efficacy of the proposed method by comparing it with state-of-the-art denoising methods using various examples.

Learning a Metric for Multimodal Medical Image Registration without Supervision Based on Cycle Constraints

Deep learning based medical image registration remains very difficult and often fails to improve over its classical counterparts where comprehensive supervision is not available, in particular for large transformations-including rigid alignment. The use of unsupervised, metric-based registration networks has become popular, but so far no universally applicable similarity metric is available for multimodal medical registration, requiring a trade-off between local contrast-invariant edge features or more global statistical metrics. In this work, we aim to improve over the use of handcrafted metric-based losses. We propose to use synthetic three-way (triangular) cycles that for each pair of images comprise two multimodal transformations to be estimated and one known synthetic monomodal transform. Additionally, we present a robust method for estimating large rigid transformations that is differentiable in end-to-end learning. By minimising the cycle discrepancy and adapting the synthetic transformation to be close to the real geometric difference of the image pairs during training, we successfully tackle intra-patient abdominal CT-MRI registration and reach performance on par with state-of-the-art metric-supervision and classic methods. Cyclic constraints enable the learning of cross-modality features that excel at accurate anatomical alignment of abdominal CT and MRI scans.

From Tokenization to Self-Supervision: Building a High-Performance Information Extraction System for Chemical Reactions in Patents

Chemical reactions and experimental conditions are fundamental information for chemical research and pharmaceutical applications. However, the latest information of chemical reactions is usually embedded in the free text of patents. The rapidly accumulating chemical patents urge automatic tools based on natural language processing (NLP) techniques for efficient and accurate information extraction. This work describes the participation of the Melax Tech team in the CLEF 2020—ChEMU Task of Chemical Reaction Extraction from Patent. The task consisted of two subtasks: (1) named entity recognition to identify compounds and different semantic roles in the chemical reaction and (2) event extraction to identify event triggers of chemical reaction and their relations with the semantic roles recognized in subtask 1. To build an end-to-end system with high performance, multiple strategies tailored to chemical patents were applied and evaluated, ranging from optimizing the tokenization, pre-training patent language models based on self-supervision, to domain knowledge-based rules. Our hybrid approaches combining different strategies achieved state-of-the-art results in both subtasks, with the top-ranked F1 of 0.957 for entity recognition and the top-ranked F1 of 0.9536 for event extraction, indicating that the proposed approaches are promising.

4S-DT: Self-Supervised Super Sample Decomposition for Transfer Learning With Application to COVID-19 Detection

Due to the high availability of large-scale annotated image datasets, knowledge transfer from pretrained models showed outstanding performance in medical image classification. However, building a robust image classification model for datasets with data irregularity or imbalanced classes can be a very challenging task, especially in the medical imaging domain. In this article, we propose a novel deep convolutional neural network, which we called self-supervised super sample decomposition for transfer learning (4S-DT) model. The 4S-DT encourages a coarse-to-fine transfer learning from large-scale image recognition tasks to a specific chest X-ray image classification task using a generic self-supervised sample decomposition approach. Our main contribution is a novel self-supervised learning mechanism guided by a super sample decomposition of unlabeled chest X-ray images. 4S-DT helps in improving the robustness of knowledge transformation via a downstream learning strategy with a class-decomposition (CD) layer to simplify the local structure of the data. The 4S-DT can deal with any irregularities in the image dataset by investigating its class boundaries using a downstream CD mechanism. We used 50000 unlabeled chest X-ray images to achieve our coarse-to-fine transfer learning with an application to COVID-19 detection, as an exemplar. The 4S-DT has achieved a high accuracy of 99.8% on the larger of the two datasets used in the experimental study and an accuracy of 97.54% on the smaller dataset, which was enriched by augmented images, out of which all real COVID-19 cases were detected.

Emergent linguistic structure in artificial neural networks trained by self-supervision

This paper explores the knowledge of linguistic structure learned by large artificial neural networks, trained via self-supervision, whereby the model simply tries to predict a masked word in a given context. Human language communication is via sequences of words, but language understanding requires constructing rich hierarchical structures that are never observed explicitly. The mechanisms for this have been a prime mystery of human language acquisition, while engineering work has mainly proceeded by supervised learning on treebanks of sentences hand labeled for this latent structure. However, we demonstrate that modern deep contextual language models learn major aspects of this structure, without any explicit supervision. We develop methods for identifying linguistic hierarchical structure emergent in artificial neural networks and demonstrate that components in these models focus on syntactic grammatical relationships and anaphoric coreference. Indeed, we show that a linear transformation of learned embeddings in these models captures parse tree distances to a surprising degree, allowing approximate reconstruction of the sentence tree structures normally assumed by linguists. These results help explain why these models have brought such large improvements across many language-understanding tasks.

NON-RIGID IMAGE REGISTRATION USING SELF-SUPERVISED FULLY CONVOLUTIONAL NETWORKS WITHOUT TRAINING DATA

A novel non-rigid image registration algorithm is built upon fully convolutional networks (FCNs) to optimize and learn spatial transformations between pairs of images to be registered in a self-supervised learning framework. Different from most existing deep learning based image registration methods that learn spatial transformations from training data with known corresponding spatial transformations, our method directly estimates spatial transformations between pairs of images by maximizing an image-wise similarity metric between fixed and deformed moving images, similar to conventional image registration algorithms. The image registration is implemented in a multi-resolution image registration framework to jointly optimize and learn spatial transformations and FCNs at different spatial resolutions with deep self-supervision through typical feedforward and backpropagation computation. The proposed method has been evaluated for registering 3D structural brain magnetic resonance (MR) images and obtained better performance than state-of-the-art image registration algorithms.