Interpretable Deep Learning Model Research Articles

Deep learning uncertainty and confidence calibration for the five-class polyp classification from colonoscopy.

There are two challenges associated with the interpretability of deep learning models in medical image analysis applications that need to be addressed: confidence calibration and classification uncertainty. Confidence calibration associates the classification probability with the likelihood that it is actually correct - hence, a sample that is classified with confidence X% has a chance of X% of being correctly classified. Classification uncertainty estimates the noise present in the classification process, where such noise estimate can be used to assess the reliability of a particular classification result. Both confidence calibration and classification uncertainty are considered to be helpful in the interpretation of a classification result produced by a deep learning model, but it is unclear how much they affect classification accuracy and calibration, and how they interact. In this paper, we study the roles of confidence calibration (via post-process temperature scaling) and classification uncertainty (computed either from classification entropy or the predicted variance produced by Bayesian methods) in deep learning models. Results suggest that calibration and uncertainty improve classification interpretation and accuracy. This motivates us to propose a new Bayesian deep learning method that relies both on calibration and uncertainty to improve classification accuracy and model interpretability. Experiments are conducted on a recently proposed five-class polyp classification problem, using a data set containing 940 high-quality images of colorectal polyps, and results indicate that our proposed method holds the state-of-the-art results in terms of confidence calibration and classification accuracy.

Explainable Anatomical Shape Analysis Through Deep Hierarchical Generative Models.

Quantification of anatomical shape changes currently relies on scalar global indexes which are largely insensitive to regional or asymmetric modifications. Accurate assessment of pathology-driven anatomical remodeling is a crucial step for the diagnosis and treatment of many conditions. Deep learning approaches have recently achieved wide success in the analysis of medical images, but they lack interpretability in the feature extraction and decision processes. In this work, we propose a new interpretable deep learning model for shape analysis. In particular, we exploit deep generative networks to model a population of anatomical segmentations through a hierarchy of conditional latent variables. At the highest level of this hierarchy, a two-dimensional latent space is simultaneously optimised to discriminate distinct clinical conditions, enabling the direct visualisation of the classification space. Moreover, the anatomical variability encoded by this discriminative latent space can be visualised in the segmentation space thanks to the generative properties of the model, making the classification task transparent. This approach yielded high accuracy in the categorisation of healthy and remodelled left ventricles when tested on unseen segmentations from our own multi-centre dataset as well as in an external validation set, and on hippocampi from healthy controls and patients with Alzheimer's disease when tested on ADNI data. More importantly, it enabled the visualisation in three-dimensions of both global and regional anatomical features which better discriminate between the conditions under exam. The proposed approach scales effectively to large populations, facilitating high-throughput analysis of normal anatomy and pathology in large-scale studies of volumetric imaging.

Learning With Interpretable Structure From Gated RNN.

The interpretability of deep learning models has raised extended attention these years. It will be beneficial if we can learn an interpretable structure from deep learning models. In this article, we focus on recurrent neural networks (RNNs), especially gated RNNs whose inner mechanism is still not clearly understood. We find that finite-state automaton (FSA) that processes sequential data have a more interpretable inner mechanism according to the definition of interpretability and can be learned from RNNs as the interpretable structure. We propose two methods to learn FSA from RNN based on two different clustering methods. With the learned FSA and via experiments on artificial and real data sets, we find that FSA is more trustable than the RNN from which it learned, which gives FSA a chance to substitute RNNs in applications involving humans' lives or dangerous facilities. Besides, we analyze how the number of gates affects the performance of RNN. Our result suggests that gate in RNN is important but the less the better, which could be a guidance to design other RNNs. Finally, we observe that the FSA learned from RNN gives semantic aggregated states, and its transition graph shows us a very interesting vision of how RNNs intrinsically handle text classification tasks.

Contrastive-Regulated CNN in the Complex Domain: A Method to Learn Physical Scattering Signatures From Flexible PolSAR Images

Single- and dual-polarimetric synthetic aperture radar (SAR) images provide very limited capabilities to interpret physical radar signatures. For generality and simplicity, we call single-polarimetric, dual-polarimetric, and fully polarimetric SAR (PolSAR) images flexible PolSAR images. In order to sufficiently extract physical scattering signatures from this kind of data and explore the potentials of different polarization modes on this task, this paper proposes a contrastive-regulated convolutional neural network (CNN) in the complex domain, attempting to learn a physically interpretable deep learning model directly from the original backscattered data. To achieve a better deep model containing physically interpretable parameters, the objective cost is compared to and selected from several commonly used loss functions in the complex form. The required ground-truth labels are generated automatically according to Cloude and Pottier’s H-alpha division plane, which significantly reduces intensive labor cost and transfers this method to an unsupervised learning mechanism. The boundaries between different scattering signatures, however, sometimes show an erroneous separation. With the aim of aggregating intra-class instances and alienating inter-class instances, meanwhile, a complex-valued contrastive regularization term is computed mathematically and is added to the objective cost by a tradeoff factor. Moreover, data augmentation is applied to relieve the side effects caused by data imbalance. Finally, we performed experiments on German Aerospace Center’s (DLR)’s L-band, high-resolution (HR), and airborne F-SAR data. Our results demonstrate the possibility of extracting physical scattering signatures from flexible PolSAR images. Physically interpretable potentials of SAR images with different polarization modes are analyzed, and we conclude with physical signature identification.

11 An Interpretable Deep Learning Model for the Prevention of Self-Harm and Suicide

Over 500,000 patients in the United States present to EDs for self-harm or suicide attempts annually, and these patients are at high risk for completed suicide. EDs are therefore a potential venue for the delivery of lifesaving interventions to prevent suicide but identifying patients at imminent risk for serious self-harm is difficult. Our objective was to develop an accurate, interpretable predictive model to identify patients at high risk for self-harm and suicide attempts, and to identify opportunities for early intervention in the ED.

Deciphering regulatory DNA sequences and noncoding genetic variants using neural network models of massively parallel reporter assays.

The relationship between noncoding DNA sequence and gene expression is not well-understood. Massively parallel reporter assays (MPRAs), which quantify the regulatory activity of large libraries of DNA sequences in parallel, are a powerful approach to characterize this relationship. We present MPRA-DragoNN, a convolutional neural network (CNN)-based framework to predict and interpret the regulatory activity of DNA sequences as measured by MPRAs. While our method is generally applicable to a variety of MPRA designs, here we trained our model on the Sharpr-MPRA dataset that measures the activity of ∼500,000 constructs tiling 15,720 regulatory regions in human K562 and HepG2 cell lines. MPRA-DragoNN predictions were moderately correlated (Spearman ρ = 0.28) with measured activity and were within range of replicate concordance of the assay. State-of-the-art model interpretation methods revealed high-resolution predictive regulatory sequence features that overlapped transcription factor (TF) binding motifs. We used the model to investigate the cell type and chromatin state preferences of predictive TF motifs. We explored the ability of our model to predict the allelic effects of regulatory variants in an independent MPRA experiment and fine map putative functional SNPs in loci associated with lipid traits. Our results suggest that interpretable deep learning models trained on MPRA data have the potential to reveal meaningful patterns in regulatory DNA sequences and prioritize regulatory genetic variants, especially as larger, higher-quality datasets are produced.

DeepSOFA: A Continuous Acuity Score for Critically Ill Patients using Clinically Interpretable Deep Learning

Traditional methods for assessing illness severity and predicting in-hospital mortality among critically ill patients require time-consuming, error-prone calculations using static variable thresholds. These methods do not capitalize on the emerging availability of streaming electronic health record data or capture time-sensitive individual physiological patterns, a critical task in the intensive care unit. We propose a novel acuity score framework (DeepSOFA) that leverages temporal measurements and interpretable deep learning models to assess illness severity at any point during an ICU stay. We compare DeepSOFA with SOFA (Sequential Organ Failure Assessment) baseline models using the same model inputs and find that at any point during an ICU admission, DeepSOFA yields significantly more accurate predictions of in-hospital mortality. A DeepSOFA model developed in a public database and validated in a single institutional cohort had a mean AUC for the entire ICU stay of 0.90 (95% CI 0.90–0.91) compared with baseline SOFA models with mean AUC 0.79 (95% CI 0.79–0.80) and 0.85 (95% CI 0.85–0.86). Deep models are well-suited to identify ICU patients in need of life-saving interventions prior to the occurrence of an unexpected adverse event and inform shared decision-making processes among patients, providers, and families regarding goals of care and optimal resource utilization.

How AI Plays Its Tricks: Interpreting the Superior Performance of Deep Learning-Based Approach in Predicting Healthcare Costs

Artificial Intelligence, especially deep learning based approaches, has been criticized for its \black-box approach in recent years. The lack of interpretability of their performance makes people uncomfortable when applying these models in critical situations. In this study, we aim to contribute to the interpretability of the deep learning model in predicting healthcare costs. We propose to use long short-term memory based recurrent neural network to incorporate the sequential information for more accurate healthcare cost predictions. We first compare the performance of our deep learning approach with traditional machine learning methods including linear regression, lasso regression, ridge regression, and random forest. Among all the methods, deep learning shows the best performance. The superior performance of the deep learning model is further confirmed in subgroup analyses of patients. We then propose a novel interpretation method to examine how the deep learning model performs differently from other methods when facing fluctuations in the monthly costs. We find that while most traditional prediction models are getting worse with greater fluctuation in the data, the deep learning model can incorporate the fluctuation information and gain in prediction accuracy. Our work makes important contributions to the interpretability of deep learning models for more accurate prediction of healthcare costs.

Discovering epistatic feature interactions from neural network models of regulatory DNA sequences.

MotivationTranscription factors bind regulatory DNA sequences in a combinatorial manner to modulate gene expression. Deep neural networks (DNNs) can learn the cis-regulatory grammars encoded in regulatory DNA sequences associated with transcription factor binding and chromatin accessibility. Several feature attribution methods have been developed for estimating the predictive importance of individual features (nucleotides or motifs) in any input DNA sequence to its associated output prediction from a DNN model. However, these methods do not reveal higher-order feature interactions encoded by the models.ResultsWe present a new method called Deep Feature Interaction Maps (DFIM) to efficiently estimate interactions between all pairs of features in any input DNA sequence. DFIM accurately identifies ground truth motif interactions embedded in simulated regulatory DNA sequences. DFIM identifies synergistic interactions between GATA1 and TAL1 motifs from in vivo TF binding models. DFIM reveals epistatic interactions involving nucleotides flanking the core motif of the Cbf1 TF in yeast from in vitro TF binding models. We also apply DFIM to regulatory sequence models of in vivo chromatin accessibility to reveal interactions between regulatory genetic variants and proximal motifs of target TFs as validated by TF binding quantitative trait loci. Our approach makes significant strides in improving the interpretability of deep learning models for genomics.Availability and implementationCode is available at: https://github.com/kundajelab/dfim.Supplementary information Supplementary data are available at Bioinformatics online.

Interpretable Representation Learning for Healthcare via Capturing Disease Progression through Time.

Various deep learning models have recently been applied to predictive modeling of Electronic Health Records (EHR). In medical claims data, which is a particular type of EHR data, each patient is represented as a sequence of temporally ordered irregularly sampled visits to health providers, where each visit is recorded as an unordered set of medical codes specifying patient's diagnosis and treatment provided during the visit. Based on the observation that different patient conditions have different temporal progression patterns, in this paper we propose a novel interpretable deep learning model, called Timeline. The main novelty of Timeline is that it has a mechanism that learns time decay factors for every medical code. This allows the Timeline to learn that chronic conditions have a longer lasting impact on future visits than acute conditions. Timeline also has an attention mechanism that improves vector embeddings of visits. By analyzing the attention weights and disease progression functions of Timeline, it is possible to interpret the predictions and understand how risks of future visits change over time. We evaluated Timeline on two large-scale real world data sets. The specific task was to predict what is the primary diagnosis category for the next hospital visit given previous visits. Our results show that Timeline has higher accuracy than the state of the art deep learning models based on RNN. In addition, we demonstrate that time decay factors and attentions learned by Timeline are in accord with the medical knowledge and that Timeline can provide a useful insight into its predictions.

Interpretable Predictions of Clinical Outcomes with An Attention-based Recurrent Neural Network.

The increasing accumulation of healthcare data provides researchers with ample opportunities to build machine learning approaches for clinical decision support and to improve the quality of health care. Several studies have developed conventional machine learning approaches that rely heavily on manual feature engineering and result in task-specific models for health care. In contrast, healthcare researchers have begun to use deep learning, which has emerged as a revolutionary machine learning technique that obviates manual feature engineering but still achieves impressive results in research fields such as image classification. However, few of them have addressed the lack of the interpretability of deep learning models although interpretability is essential for the successful adoption of machine learning approaches by healthcare communities. In addition, the unique characteristics of healthcare data such as high dimensionality and temporal dependencies pose challenges for building models on healthcare data. To address these challenges, we develop a gated recurrent unit-based recurrent neural network with hierarchical attention for mortality prediction, and then, using the diagnostic codes from the Medical Information Mart for Intensive Care, we evaluate the model. We find that the prediction accuracy of the model outperforms baseline models and demonstrate the interpretability of the model in visualizations.