Spain

In this presentation, we quantify the uncertainty of deep neural networks (DNNs) for the task of Chest X-Ray (CXR) image classification. We investigate uncertainties of several commonly used DNN architectures including ResNet, ResNeXt, DenseNet and SENet. We propose an uncertainty based strategy and analyze the impact of this strategy on the classifier performance. Results show that utilizing uncertainty information may improve DNN performance for some metrics and observations.

ECAI 2020

Uncertainty Quantification in Chest X-Ray Image Classification using Bayesian Deep Neural Networks

ECAI2020 featured a full programme of technical papers, as well as many AI-related events of interest for researchers, students and all attendants who are interested on contemporary AI, including:




**An extensive programme of workshops, tutorials, and invited speakers**

**The Prestigious Applications of Intelligent Systems conference (PAIS)**

**The Frontiers in AI track sessions**

**The Lab To Market event**

**The Women in AI time**

**The EU Challenges forum**

… and many more!

ECAI 2020 featured a full programme of technical papers, as well as many AI-related events of interest for researchers, students and all attendants who are interested on contemporary AI.

technical paper

The use of artificial pancreas systems is becoming increasingly popular amongst patients with type 1 diabetes due to its effective ability to allow the patient better control of patient's own blood glucose levels compared to other more standard treatments. But what if the glucose monitor stops transmitting for an unpredictable period of time while the pump is still working? How should the pump react to guarantee the safety of the patients? In this presentation, we will discuss some potential failure mode strategies that we have explored and their corresponding results.

In Silico Comparison of Continuous Glucose Monitor Failure Mode Strategies for and Artificial Pancreas

"Knowledge Discovery, Processing and Representation. Since in our modern society the healthcare system plays an important role and has a major impact in our daily lives, it lies at hand to apply the aforementioned methods in order to discover relevant patterns in healthcare databases and then to represent them in a way
which supports reasoning, decision making, and communication. We approach this task by using two complementary directions, which are then interlinked. On the one hand we make use of the graphical representation capabilities of Formal Concept Analysis (FCA) and its powerful algorithms for conceptual knowledge discovery and
processing. On the other, we use graph databases as a complementary visualization method of the extracted knowledge patterns. We exemplify this approach on a particular medical dataset, highlighting a 3D representation of conceptual hierarchies by using virtual reality (VR)."

Knowledge Discovery and Visualization in Healthcare Datasets using Formal Concept Analysis and Graph Databases

"Most advanced machine learning techniques, such as Artificial Neural Networks (ANNs), are black boxes and therefore not interpretable for humans. 
A way of explaining an ANN is visualizing the internal feature representations of its hidden layers (neural embeddings). However, interpreting these visualizations is still difficult. 
We therefore present InterVENE: an approach that visualizes neural embeddings and interactively explains this visualization, aiming for knowledge extraction and network interpretation. We project neural embeddings in a 2-dimensional scatter plot, where users can interactively select two subsets of data instances in this visualization. Subsequently, a personalized decision tree is trained to distinguish these two sets, thus explaining the difference between the two sets.
We apply InterVENE to a medical case study where interpretability of decision support is critical: outcome prediction of comatose patients. Our experiments confirm that InterVENE can successfully extract knowledge from an ANN, and give both domain experts and machine learning experts insight into the behaviour of an ANN. Furthermore, InterVENE's explanations about outcome prediction of comatose patients seem plausible when compared to existing neurological domain knowledge."

Interactive Explanations of Internal Representations of Neural Network Layers: An Exploratory Study on Outcome Prediction of Comatose Patients

Fracture detection has been a long-standing paradigm on the medical imaging community. Many algorithms and systems have been presented to accurately detect and classify images in terms of the presence and absence of fractures in different parts of the body. While these solutions are capable of obtaining results which even surpass human scores, few efforts have been dedicated to evaluate how these systems can be embedded in the clinicians and radiologists working pipeline. Moreover, the reports that are included with the radiography could also provide key information regarding the nature and the severity of the fracture. In this paper, we present our first findings towards assessing how computer vision, natural language processing and other systems could be correctly embedded in the clinicians' pathway to better aid on the fracture detection task. We present some initial experimental results using publicly available fracture datasets along with a handful of data provided by the National Healthcare System from the United Kingdom in a research initiative call. Results show that there is a high likelihood of applying transfer learning from different existing and pre-trained models to the new records provided in the challenge, and that there are various ways in which these techniques can be embedded along the clinicians' pathway.

Assessing the clinicians' pathway to embed artificial intelligence for assisted diagnostics of fracture detection

"The management of type 1 diabetes mellitus (T1DM) is a burdensome life-long task. In fact, T1DM individuals are request to perform every day tens of actions to adapt the insulin therapy, aimed at maintaining the blood glucose (BG) concentration as much as possible into a safe range coping with the day-to-day variability of their life style. The recent availability of continuous glucose monitoring (CGM) devices and other low-cost wearable sensors to track important vital and activity signals, is stimulating the development of decision support systems to lower this burden. Modern deep learning models, trained using rich amount of information, are a suitable and effective instrument for such purpose, especially if used to predict future BG values. However, the high accuracy of deep learning approaches is often obtained at the expense of less interpretability.

To surpass this limit, in this work we propose a new deep learning method for BG prediction based on a personalized bidirectional long short-term memory (LSTM) equipped with a tool that enables its interpretability. The OhioT1DM Dataset was used to develop a model targeting future BG at 30 and 60 minute prediction horizons (PH). The accuracy of model predictions was evaluated in terms of root mean square error (RMSE), mean absolute error (MAE), and the time gained (TG) to anticipate the actual glucose concentration.

The obtained results show fairly good prediction accuracy (for PH = 30/60 min): RMSE = 20.20/34.19 mg/dl, MAE = 14.74/25.98 mg/dl, and TG = 9.17/18.33 min.
Moreover, we showed, in a representative case, that our algorithm is able to preserve the physiological meaning of the considered inputs. 

In conclusion, we built a model able to provide reliable glucose performance ensuring the interpretability of its output. Future work will assess model performance against other competitive strategies. "

A Personalized and Interpretable Deep Learning Based Approach to Predict Blood Glucose Concentration inType 1 Diabetes

A machine learning-based method for blood glucose level prediction thirty and sixty minutes in advance based on highly multiclass classification (as opposed to the more traditional regres- sion approach) is proposed. An advantage of this approach is the possibility of modelling and visualising the uncertainty of a predic- tion across the entire range of blood glucose levels without paramet- ric assumptions such as normality. To demonstrate the approach, a long-short term memory-based neural network classifier is used in conjunction with a blood glucose-specific data preprocessing tech- nique (risk domain transform) to train a set of models and generate predictions for the 2018 and 2020 Blood Glucose Level Prediction Competition datasets. Numeric accuracy results are reported along with examples of the uncertainty visualisation possible using this technique.

Neural Multi-class Classification Approach to Blood Glucose Level Forecasting with Prediction Uncertainty Visualisation

For people with type 1 diabetes (T1D), the beta cells in pancreas are destroyed and can not produce insulin which is the only hormone that response to high glucose concentrations. So, people with T1D must administer exogenous insulin to maintain their BGC within an euglycemic range. Hyperglycemia (high BGC) and hypoglycemia (low BGC) can occur because of poor BGC management. Using recursively identified models to predict the future BGC values opens novel possibilities for improving the BGC regulation performance by adjusting the dose of insulin infusion, taking rescue carbohydrates, or both. In this paper, a latent variable (LV) based multivariable statistical modeling approach is applied to model BGC dynamics and forecast future BGCs. The model is evaluated with the Ohio T1DM dataset that contains BGCs from continuous glucose monitoring (CGM) sensors, basal and bolus insulin information from insulin pump, and additional information from wristbands or reported by the subject. The average RMSE of 6 subjects is 19.37 mg/dL for 30-min-ahead prediction and 32.59 mg/dL for 60-min-ahead prediction. The results indicate that the LV-based model can predict future BGC values with high prediction accuracy.

Prediction of Glucose Levels for People with Type 1 Diabetes using Latent Variable based Model

In this presentation, we show our results for the Blood Glucose Level Prediction Challenge for the Ohio2020 dataset. We have used four variants of genetic programming to build white-box models for predicting 30 minutes and 60 minutes ahead. The results are compared to classical methods including multivariate linear regression, random forests, as well as two types of ARIMA models. Notably, we have included future values of bolus and basal into some of the models because we assume that these values can be controlled. Additionally, we have used a convolution filter to smooth the information in the bolus volume feature. We find that overall tree-based GP per-forms well and better than multi-variate linear regression and random forest, while ARIMA models performed worst on the here analyzed data.

Analysis of the performance of Genetic Programming on the Blood Glucose Level Prediction Challenge 2020

"An essential part of this work is to provide a data-driven
model for predicting blood glucose levels that will help to warn the
patient about a potential hypo- or hyper-glycemic event in an easyto-
manage and discreet way."

A Deep Learning Approach for Blood Glucose Prediction of Type 1 Diabetes

"Real-time forecasting of blood glucose (BG) levels has
the potential to drastically improve management of Type 1 Diabetes,
a widespread chronic disease affecting the metabolic system. Most
notably, if hypo or hyperglycemia episodes (i.e. glycemic excursion
below or above a safe range) could be accurately predicted, then the
patient could be timely warned, thus enabling proactive countermeasures
to avoid these dangerous conditions. In this work, a novel personalized
algorithm for the real-time forecasting of BG is developed
by combining the output of a shallow feed forward neural network
with an error imputation module composed by an ensemble of trees.
Past glucose readings as well as insulin, meals and work/sleep time
information are carefully handled to train and boost the prediction
performance of the algorithm. The root mean square error over the 6
subjects achieves a mean value of 18.63 mg/dL and 32.27 mg/dL for
30- and 60-minute prediction horizon respectively."

Uncertainty Quantification in Chest X-Ray Image Classification using Bayesian Deep Neural Networks

Downloads

Next from ECAI 2020

In Silico Comparison of Continuous Glucose Monitor Failure Mode Strategies for and Artificial Pancreas

Similar lecture

Knowledge Discovery and Visualization in Healthcare Datasets using Formal Concept Analysis and Graph Databases

Uncertainty Quantification in Chest X-Ray Image Classification using Bayesian Deep Neural Networks

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from ECAI 2020

In Silico Comparison of Continuous Glucose Monitor Failure Mode Strategies for and Artificial Pancreas

Similar lecture

Knowledge Discovery and Visualization in Healthcare Datasets using Formal Concept Analysis and Graph Databases

Downloads