Image source: istock.com/dragana991

NHS Greater Glasgow and Clyde (NHSGGC) has launched a clinical investigation of how machine learning can be used in workflows to support the care of people with chronic obstructive pulmonary disease (COPD).

It is working with Lenus Health on the Dynamic-AI 12-month feasibility study on how models based on machine learning could identify patients at higher risk of adverse events.

This could support proactive interventions and help to reduce the number of emergency hospital admissions.

Lenus Health said this is the first project to operationalise predictive AI in routine direct care of chronic conditions.

It follows extensive co-design efforts with patients and clinicians to develop a technically viable and clinically explainable way of delivering risk scores derived from machine learning derived into existing care pathways.

This is an incredibly exciting project. Its the first time were bringing together predictive AI insight for COPD into live clinical practice, said Dr Chris Carlin, consultant respiratory physician at NHSGGC, who is leading the investigation. With the ageing population and rising prevalence and complexity of long term conditions, clinicians are overwhelmed with data that they dont have the capacity to review.

We need to deploy assistive technologies to provide us with prioritised insights from patient data. These have the potential to give us back time to focus on patient-clinician human interactions and allows us to optimise preventative management to improve patient outcomes and quality of life rather than continue to firefight with unsustainable reactive unscheduled care.

Supported by a 1.2 million NHS Artificial Intelligence in Health and Care Award in 2021 an NHS AI Lab programme Lenus Healths team of data scientists and engineers have developed four machine learning models to proactively identify patients with COPD who are at risk of adverse events and provide actionable insights to improve care quality.

The proprietary AI algorithms are UKCA (UK Conformity Assessed) marked and were trained using close to one million data points from historical electronic health records from a de-identified cohort of more than 55,000 patients with COPD who live in the NHSGGC area.

Lenus Health uses more than 80 data points to support the delivery or risk scores, significantly more than in a traditional rule based system, which are known to cause numerous false alarms and lead to clinicians experiencing alarm fatigue.

Rule based systems are static whereas machine learning is much more robust in the context of routine care, Carlin said.

Clinical care teams will be provided with actionable insights from the models to use in multi-disciplinary team (MDT) reviews. By identifying high risk patients, they can be offered pro-active, preventative care to avoid the COPD symptom flare ups that currently cause one in eight emergency hospital admissions.

One of the key things we hope this will tell us is which patients are at risk of adverse outcome so we can provide anticipatory care, Carlin said. We will be able to discuss with patients the level of care they want in advance rather than in an emergency situation which is much more pressured.

The new study builds on a previous collaboration between Lenus Health and NHSGGC, which produced a digital service model for supported self-management of COPD patients.

Patients currently using the digital COPD service at NHSGGC will have the option to consent to take part in the AI study, which has been approved by the Medicines and Healthcare Products Regulatory Agency (MHRA).

See more here:
NHS Greater Glasgow and Clyde trials machine learning in care of ... - UKAuthority.com

Anyone aiming to work with deep learning and artificial intelligence systems on-the-go with a notebook or small form factor PC is about to get an option to supercharge their hardware, courtesy of a pocket-sized external GPU from ADLINK: the Pocket AI, based on an NVIDIA RTX A500 add-in board.

"Pocket AI offers the ultimate in flexibility and reliability on the move," ADLINK claims of its impending hardware launch. "This portable, plug and play AI accelerator delivers a perfect power/performance balance from the NVIDIA RTX GPU. Pocket AI is the perfect device for AI developers, professional graphics users and embedded industrial applications for boosting productivity by improve the work efficiency."

The compact device takes advantage of the ability to use PCI Express lanes over a Thunderbolt USB Type-C connector to interface devices without the room for a full-size PCIe graphics card like notebooks to one of NVIDIA's more powerful examples.

In the case of the launch model, that's an NVIDIA RTX A500 graphics card, based on the Ampere GA107 architecture, which offers 2,048 CUDA cores, 64 Tensor cores, and 16 RT cores. All told, that's equivalent to a claimed 6.54 tera-floating point operations per second (TFLOPS) of FP32-precision compute, or 100 tera-operations per second (TOPS) of dense INT8 performance.

The idea behind the device: to act as a simple plug-and-play accelerator for deep learning and artificial intelligence workloads, as well as graphically-demanding tasks including ray tracing. The device is powered by USB Power Delivery (USB PD) on a second Type-C port, measures just 10672mm (around 4.172.83"), and weighs 250g (around 8.8oz) making it easily portable. The only catch: just 4GB of GDDR6 memory, meaning that the device may struggle fitting larger models into RAM.

More information on the device is available on the ADLINK website, though the company has yet to reveal pricing but has confirmed plans to open pre-orders this month, with shipping due to commence in June.

Freelance journalist, technical author, hacker, tinkerer, erstwhile sysadmin. For hire: freelance@halfacree.co.uk.

Continued here:
ADLINK Wants to Put a Powerful GPGPU in Your Pocket for On-the-Go Machine Learning Acceleration - Hackster.io

In this section we describe the clinical data set, data preprocessing, feature selection process, classifiers used, and design for the machine learning experiments.

The data we study is a collection of seven clinical studies available via the NCBI Gene Expression Omnibus, Series GSE73072. The details of these studies can be found in1, but we briefly summarize here and in Fig.1.

Each of the seven studies enrolled individuals to be infected with one of four viruses associated with a common respiratory infection. Studies DEE2-DEE5 challenged participants with H1N1 or H3N2. Studies Duke and UVA challenged participants with HRV, while DEE1 challenged individuals with RSV.

In all cases, individuals had blood samples taken at regular intervals every 412h both prior to and after infection; see Fig.1 for details. Specific time points are measured as hours since infection and vary by study. In total, 148 human subjects were involved with approximately 20 sampled time points per person. Blood samples were run through undirected microarray assays. CEL data files available via GEO were read and processed using RMA (Robust Multi-array Average) normalization through use of several Bioconductor packages23 producing expression values across 22,277 microarray probes.

To address the time of infection question, we separate the training and test samples into 9 bins in time post-inoculation, each with a categorical label; see Fig.1. The first six categories correspond to disjoint 8-h intervals in the first 2days after inoculation, and the last three categories are disjoint 24-h intervals from hours 48 to 120h. In addition to this 9-class classification problem, we also studied a relaxed binary prediction problem of whether a subject belongs to the early phase of infection (time of inoculation (le ) 48h) or later phase (time of infection > 48h). Results for this binary classification are inferred from the 9-class problem, i.e., if a classified label is associated to a time in the first 2days, it is considered correctly labeled.

After the data is processed, we apply the following general pipeline for each of the 14 experiments enumerated in Fig.2 (top panel):

Partition the data into training and testing sets based on the classification experiment.

Normalize the data to correct for batch effects seen between subjects (e.g., using the linear batch normalization routine in the limma package24).

Identify comprehensive sets of predictive features using the Iterative Feature Removal (IFR) approach6, which aims to extract all discriminatory features in high-dimensional data with repeated application of sparse linear classifiers such as Sparse Support Vector Machines (SSVM).

Identify network architectures and other hyperparameters for Artificial Neural Networks (ANN) and Centroid Encoder (CE) by utilizing a five-fold cross-validation experiment on the training data.

Evaluate the features identified from the step 3 on the test data. This is done by training and evaluating a new model using the selected features with a leave-one-subject-out cross validation scheme on the test study. The metric used for evaluation is BSR; throughout this study we utilize BSR as a balanced representation of performance accounting for imbalanced class sizes while being easy to interpret.

For each of the training sets illustrated in Fig.2 (top panel; stripes), features are selected using the IFR algorithm, with an SSVM classifier. This is done separately for all pairwise combinations of categorical labels (time bins); a 9-class experiment leads to 9-choose-2 = 36 pairwise combinations. So, for each of these 36 combinations of time bins, features are selected using the following steps.

First, the input data to IFR algorithm is partitioned into training and validation set. Next, sets of features that produce high accuracy on the validation set are selected iteratively. In each iteration, features that have previously been selected are masked out, so that theyre not used again. The feature selection is halted once the predictive rates on the validation data drops below a specified threshold. This results in one feature-set for a particular training-validation set of the input data.

Next, more training-validation partitions are repeatedly sampled, and the feature selection, as described above, is repeated for each partition-set; this results in a different feature-set for each new partition-set. Then, these different feature-sets are combined by applying a set union operation and the frequency of each individual feature is tracked if they are discovered in multiple feature-sets. The feature frequency is used to rank the features; the more a particular feature is discovered, the more important it is.

The size of this combined feature-set, although about 520% of the original feature-set size, is still often large for classification, so a last step we reduce the size of this feature-set. This is done by performing a grid-search using a linear SVM (without sparsity penalty) on the training data, taking the top n features, ranked by frequency, which maximize the average of true positive rates on every class, or BSR. Once the features have been selected, we perform a more detailed leave-one-subject-out classification for the experiments described in the Results and visualized in Fig.2 using the classifiers described in Methods section.

Feature selection produced 36 distinct feature-sets, coming from all distinct choices of two time bins from the nine labels possible. To address the question of commonality or importance of features selected on a time bin for a specific pathway, we implemented a heuristic scoring system. For a fixed time bin (say, bin1) and a fixed feature-set (say, bin1_vs_bin2; quantities summarized in Table 2) the associated collection of features was referenced against the GSEA MSigDB. This database includes both canonical pathways and gene sets as the result of data miningwe refer to anything in this collection generically as a gene set. A score for each MSigDB gene set was assigned for a given feature-set (bin1_vs_bin2) based on the ratio of features in the feature-set which appear in the gene set. For instance, a score of 0.5 for hypothetical GENE_SET_A for feature-set bin1_vs_bin2 represents the fact that 50% of the features in GENE_SET_A are present in bin1_vs_bin2.

A score for pathway on a time bin by itself was defined as the sum of the scores for that pathway on all feature-sets related to it. Continuing the example, a score for GENE_SET_A on bin1 would be the sum of the scores for GENE_SET_A for feature-set bin1_vs_bin2, bin1_vs_bin3, all the way up to bin1_vs_bin9, with equal weighting.

Certainly, there are several subtle statistical and combinatorial questions relating to this procedure. Direct comparison of pathways and gene sets is challenging due their overlapping nature (features may belong to multiple gene sets). The number of features associated with a gene set can vary anywhere from under 10, to over 1000 and may complicate a scoring system based on percentage overlap, such as ours. Attempting to use a mathematically or statistically rigorous procedure to account for these, and other potential factors is a worthy exercise, but we believe our heuristic is sufficient for an explainable high-level summary of the composition of the feature-sets found.

In this section we describe the classifiers and how they are applied for the classification task. We also describe how the feature-sets are adapted to different classifiers.

After feature selection, we evaluate the features on test sets based on successful classification in the nine time bins. For each experiment shown in Fig.1, we use the feature-sets extracted on its training set and evaluate the models using leave-one-subject-out cross validation on the test set. Each experiment is repeated 25 times to capture variability. For the binary classifiersSSVM and linear SVMwe used a multiclass method, with each of its ({9 atopwithdelims ()2}) pairwise models using respective feature-sets. On the other hand, we used a single classification model for ANN and CE because these models can handle multiple classes. The feature-set for these models are created by taking a union of ({9 atopwithdelims ()2} = 36) pairwise feature-sets.

Balanced Success Rate (BSR) Throughout the Results section, we report predictive power in terms of BSR. This is a simple average of true positive rates for each of the categories. The BSR serves as a simple, interpretable metric especially when working with imbalanced data sets and gives a holistic view of classification performance that easily generalizes to multiclass problems. For example, if true positive rates in a 3-class problem were (TPR_1 = 95%), (TPR_2 = 50%), and (TPR_3 = 65%), the BSR for the multiclass problem would be ((TPR_1 + TPR_2 + TPR_3)/3 = 70%).

We implement a pairwise model (or one-vs-one model) for training and classification to extend the binary classifiers described below (ANN and CE do not require these). For a data set with c unique classes, c-choose-2 models are built using the relevant subsets of the data. Learned model parameters and features selected for each model are stored and later used when discriminatory features are needed in the test phase.

After training, classification is done by a simple voting scheme: a new sample is classified by all c-choose-2 classifiers and assigned the label that had the plurality of the vote. If a tie occurs, the class is decided by an unbiased coin flip between the winning labels. In a nine-class problem, this corresponds to 36 classifiers and feature-sets being selected.

Linear SVM For a plain linear SVM model, the implementation in the scikit-learn package in Python was used25. While scikit-learn also has built-in support to extend this binary classifier to multiclass problems, either by one-vs-one or one-vs-all approaches, we only use it for binary classification problems, or for binary sub-problems of a one-vs-one scheme for a multiclass problem. The optimization problem was introduced by26 and requires the solution to

$$begin{aligned} begin{aligned} textrm{min}_{w,b} ;&||w||_2^2 quad text {subject to} \&y^i ( w cdot x^i - b ) ge 1, quad text { for all } i end{aligned} end{aligned}$$

(1)

where (y^i) represent class labels assigned to (pm 1), (x^i) represent vector samples, w represents the weight vector and b represents a bias (a scalar shift). This approach has seen widespread use and success in biological feature extraction27,28.

Sparse SVM (SSVM) The SSVM problem replaces the 2-norm in the objective of equation 1 with a 1-norm, which is understood to promote sparsity (many zero coefficients) in the coefficient vector ({textbf{w}}). This allows one to ignore those features and is our primary tool for feature selection when coupled with Iterated Feature Removal6. Arbitrary p-norm SVM were introduced in29 and (ell _1)-norm sparse SVM were further developed for feature selection in6,30,31.

After a standard one-hot encoding scheme, inherently multiclass methods (here: neural networks) do not need to be adapted to handle a multiclass problem as with linear methods, nor is there a straightforward way to encode the use of time-dependent features in passing new data forward through the neural network; this would be begging the (time of infection) question. Instead, for these methods, we simply take the union of all pairwise features built to classify pairs of time bins, then allow the multiclass algorithm to learn any necessary relationships internally. The specifics of the neural networks are described below.

Artificial Neural Networks (ANN) We apply a standard feed-forward neural network trained to learn the labels of the training data. In all the classification tasks, we used two hidden layers with 500 ReLU activation in each layer. We used the whole training set to calculate the gradient of the loss function (Cross-entropy) while updating the network parameters using Scaled Conjugate Gradient Descent (SCG); see32.

Centroid-Encoder (CE) This is a variation of an autoencoder which can be used for both visualization and classification purposes. Consider a data set with N samples and M classes. The classes denoted (C_j, j = 1, dots , M) where the indices of the data associated with class (C_j) are denoted (I_j). We define centroid of each class as (c_j=frac{1}{|C_j|}sum _{i in I_j} x_i) where (|C_j|) is the cardinality of class (C_j). Unlike autoencoder, which maps each point (x_i) to itself, CE will map each point (x_i) to its class centroid (c_j) by minimizing the following cost function over the parameter set (theta ):

$$begin{aligned} begin{aligned} {mathscr {L}}_{ce}(theta )=frac{1}{2N}sum ^M_{j=1} sum _{i in I_j}Vert c_j-f(x_i; theta ))Vert ^2_2 end{aligned} end{aligned}$$

(2)

The mapping f is composed of a dimension reducing mapping g (encoder) followed by a dimension increasing reconstruction mapping h (decoder). The output of the encoder is used as a supervised visualization tool33, and attaching another layer to map to the one-hot encoded labels and further training by fine-tuning provides a classifier. For further details, see34. In all of the classification tasks, we used three hidden layers ((500 rightarrow 100 rightarrow 500)) with ReLU activation for centroid mapping. After that we attached a classification layer with one-hot-encoding to the encoder ((500 rightarrow 100)) to learn the class label of the samples. The model parameters were updated using SCG.

Continue reading here:
Using machine learning to determine the time of exposure to ... - Nature.com

Research led by Tsinghua University, Beijing, has developed a deep learning (DL) program that can improve prognostic biomarker discovery to help patients with liver cancer.

The researchers used the tool, known as PathFinder, to show the value of a biomarker that plays a key role in liver cancer outcomes. They also hope it can be useful for finding biomarkers for different types of cancer in the future.

Tissue biomarkers are crucial for cancer diagnosis, prognosis assessment and treatment planning. However, there are few known biomarkers that are robust enough to show true analytical and clinical value, write Lingjie Kong, a senior researcher from Tsinghua University, Beijing, and colleagues in the journal Nature Machine Intelligence.

DL-based computational pathology can be used as a strategy to predict survival, but the limited interpretability and generalizability prevent acceptance in clinical practice Thus there is still a desperate need for identifying additional robust biomarkers to guide tumor diagnosis and prognosis, and to direct the research of tumor mechanism.

PathFinder is a DL-guided framework that is designed to be easy to interpret for pathologists and other healthcare professionals or researchers who are not computational experts. It uses a combination of whole slide images from patients with cancer and healthy controls with spatial information, as well as DL, to search for new biomarkers.

In this study, using liver cancer as an example, the tool showed spatial distribution of necrosis in liver cancer is strongly related to patient prognosis. This biomarker is known, but rarely used in current clinical practice.

From their findings, the research team suggested two measurements, necrosis area fraction and tumor necrosis distribution, as ways pathologists can assess spatial distribution of necrosis in liver cancer patients to improve the accuracy of prognostic predictions. They then verified these measures in the Cancer Genome Atlas Liver Hepatocellular Carcinoma dataset and the Beijing Tsinghua Changgung Hospital dataset.

By combining sparse multi-class tissue spatial distribution information of whole slide images with attribution methods, PathFinder can achieve localization, characterization and verification of potential biomarkers, while guaranteeing state-of-the-art prognostic performance, write the authors.

In this study, we did not target AI as a substitute for pathologists, but as a tool for pathologists to mine dominate biomarkers. Just as AI guides mathematical intuition, pathologists can formulate specific hypotheses based on their clinical experience, and then use PathFinder to deeply mine the connection between hypotheses-relevant information and prognosis.

More:
Deep Learning Tools Can Improve Liver Cancer Prognosis - Inside Precision Medicine

Merging physics-based simulations with artificial intelligence gains increasing importance in materials science, especially for the design of complex materials that meet technological and environmental demands. Credit: T. You, Max-Planck-Institut fr Eisenforschung GmbH

Max Planck scientists explore the possibilities of artificial intelligence in materials science and publish their review in the journal Nature Computational Science.

Advanced materials become increasingly complex due to the high requirements they have to fulfil regarding sustainability and applicability. Dierk Raabe, and colleagues reviewed the use of artificial intelligence in materials science and the untapped spaces it opens if combined with physics-based simulations. Compared to traditional simulation methods, AI has several advantages and will play a crucial role in material sciences in the future.

Advanced materials are urgently needed for everyday life, be it in high technology, mobility, infrastructure, green energy or medicine. However, traditional ways of discovering and exploring new materials encounter limits due to the complexity of chemical compositions, structures and targeted properties. Moreover, new materials should not only enable novel applications, but also include sustainable ways of producing, using and recycling them.

Researchers from the Max-Planck-Institut fr Eisenforschung (MPIE) review the status of physics-based modelling and discuss how combining these approaches with artificial intelligence can open so far untapped spaces for the design of complex materials. They published their perspective in the journal Nature Computational Science.

To meet the demands of technological and environmental challenges, ever more demanding and multifold material properties have to be considered, thus making alloys more complex in terms of composition, synthesis, processing and recycling. Changes in these parameters entail changes in their microstructure, which directly impacts materials properties. This complexity needs to be understood to enable the prediction of structures and properties of materials. Computational materials design approaches play a crucial role here.

Our means of designing new materials rely today exclusively on physics-based simulations and experiments. This approach can experience certain limits when it comes to the quantitative prediction of high-dimensional phase equilibria and particularly to the resulting non-equilibrium microstructures and properties. Moreover, many microstructure- and property-related models use simplified approximations and rely on a large number of variables. However, the question remains if and how these degrees of freedom are still capable of covering the materials complexity, explains Professor Dierk Raabe, director at MPIE and first author of the publication.

The paper compares physics-based simulations, like molecular dynamics and ab initio simulations with descriptor-based modelling and advanced artificial intelligence approaches. While physics-based simulations are often too costly to predict materials with complex compositions, the use of artificial intelligence (AI) has several advantages.

AI is capable of automatically extracting thermodynamic and microstructural features from large data sets obtained from electronic, atomistic and continuum simulations with high predictive power, says Professor Jrg Neugebauer, director at MPIE and co-author of the publication.

As the predictive power of artificial intelligence depends on the availability of large data sets, ways of overcoming this obstacle are needed. One possibility is to use active learning cycles, where machine learning models are trained with initially small subsets of labelled data. The models predictions are then screened by a labelling unit that feeds high quality data back into the pool of labelled records and the machine learning model is run again. This step-by-step approach leads to a final high-quality data set usable for accurate predictions.

There are still many open questions for the use of artificial intelligence in materials science: how to handle sparse and noisy data? How to consider interesting outliers or misfits? How to implement unwanted elemental intrusion from synthesis or recycling? However, when it comes to designing compositionally complex alloys, artificial intelligence will play a more important role in the near future, especially with the development of algorithms, and the availability of high-quality material datasets and high-performance computing resources.

Reference: Accelerating the design of compositionally complex materials via physics-informed artificial intelligence by Dierk Raabe, Jaber Rezaei Mianroodi and Jrg Neugebauer, 31 March 2023, Nature Computational Science.DOI: 10.1038/s43588-023-00412-7

The research is supported by the BigMax network of the Max Planck Society.

Excerpt from:
Merging Artificial Intelligence and Physics Simulations To Design ... - SciTechDaily