Search

The success of convolutional neural networks (CNNs) in various applications is accompanied by a significant increase in computation and parameter storage costs. Recent efforts to reduce these overheads involve pruning and compressing the weights of various layers while at the same time aiming to not sacrifice performance. In this paper, we propose a novel criterion for CNN pruning inspired by neural network interpretability: The most relevant units, i.e. weights or filters, are automatically found using their relevance scores obtained from concepts of explainable AI (XAI). By exploring this idea, we connect the lines of interpretability and model compression research. We show that our proposed method can efficiently prune CNN models in transfer-learning setups in which networks pre-trained on large corpora are adapted to specialized tasks. The method is evaluated on a broad range of computer vision datasets. Notably, our novel criterion is not only competitive or better compared to state-of-the-art pruning criteria when successive retraining is performed, but clearly outperforms these previous criteria in the resource-constrained application scenario in which the data of the task to be transferred to is very scarce and one chooses to refrain from fine-tuning. Our method is able to compress the model iteratively while maintaining or even improving accuracy. At the same time, it has a computational cost in the order of gradient computation and is comparatively simple to apply without the need for tuning hyperparameters for pruning., Comment: 25 pages + 5 supplementary pages, 13 figures, 6 tables

Deep learning has recently gained popularity in digital pathology due to its high prediction quality. However, the medical domain requires explanation and insight for a better understanding beyond standard quantitative performance evaluation. Recently, explanation methods have emerged, which are so far still rarely used in medicine. This work shows their application to generate heatmaps that allow to resolve common challenges encountered in deep learning-based digital histopathology analyses. These challenges comprise biases typically inherent to histopathology data. We study binary classification tasks of tumor tissue discrimination in publicly available haematoxylin and eosin slides of various tumor entities and investigate three types of biases: (1) biases which affect the entire dataset, (2) biases which are by chance correlated with class labels and (3) sampling biases. While standard analyses focus on patch-level evaluation, we advocate pixel-wise heatmaps, which offer a more precise and versatile diagnostic instrument and furthermore help to reveal biases in the data. This insight is shown to not only detect but also to be helpful to remove the effects of common hidden biases, which improves generalization within and across datasets. For example, we could see a trend of improved area under the receiver operating characteristic curve by 5% when reducing a labeling bias. Explanation techniques are thus demonstrated to be a helpful and highly relevant tool for the development and the deployment phases within the life cycle of real-world applications in digital pathology.

The eligibility for hormone therapy to treat breast cancer largely depends on the tumor’s estrogen receptor (ER) status. Recent studies show that the ER status correlates with morphological features found in Haematoxylin-Eosin (HE) slides. Thus, HE analysis might be sufficient for patients for whom the classifier confidently predicts the ER status and thereby obviate the need for additional examination, such as immunohistochemical (IHC) staining. Several prior works are limited by either the use of engineered features, multi-stage models that use features unspecific to HE images or a lack of explainability. To address these limitations, this work proposes an end-to-end neural network ensemble that shows state-of-the-art performance. We demonstrate that the approach also translates to the prediction of the cancer grade. Moreover, subsets can be selected from the test data for which the model can detect a positive ER status with a precision of 94% while classifying 13% of the patients. To compensate for the reduced interpretability of the model that comes along with end-to-end training, this work applies Layer-wise Relevance Propagation (LRP) to determine the relevant parts of the images a posteriori, commonly visualized as a heatmap overlayed with the input image. We found that nuclear and stromal morphology and lymphocyte infiltration play an important role in the classification of the ER status. This demonstrates that interpretable machine learning can be a vital tool for validating and generating hypotheses about morphological biomarkers.

The extent of tumor-infiltrating lymphocytes (TILs), along with immunomodulatory ligands, tumor-mutational burden and other biomarkers, has been demonstrated to be a marker of response to immune-checkpoint therapy in several cancers. Pathologists have therefore started to devise standardized visual approaches to quantify TILs for therapy prediction. However, despite successful standardization efforts visual TIL estimation is slow, with limited precision and lacks the ability to evaluate more complex properties such as TIL distribution patterns. Therefore, computational image analysis approaches are needed to provide standardized and efficient TIL quantification. Here, we discuss different automated TIL scoring approaches ranging from classical image segmentation, where cell boundaries are identified and the resulting objects classified according to shape properties, to machine learning-based approaches that directly classify cells without segmentation but rely on large amounts of training data. In contrast to conventional machine learning (ML) approaches that are often criticized for their “black-box” characteristics, we also discuss explainable machine learning. Such approaches render ML results interpretable and explain the computational decision-making process through high-resolution heatmaps that highlight TILs and cancer cells and therefore allow for quantification and plausibility checks in biomedical research and diagnostics., SCOPUS: re.j, info:eu-repo/semantics/published

Head and neck squamous cell carcinoma (HNSC) patients are at risk of suffering from both pulmonary metastases or a second squamous cell carcinoma of the lung (LUSC). Differentiating pulmonary metastases from primary lung cancers is of high clinical importance, but not possible in most cases with current diagnostics. To address this, we performed DNA methylation profiling of primary tumors and trained three different machine learning methods to distinguish metastatic HNSC from primary LUSC. We developed an artificial neural network that correctly classified 96.4% of the cases in a validation cohort of 279 patients with HNSC and LUSC as well as normal lung controls, outperforming support vector machines (95.7%) and random forests (87.8%). Prediction accuracies of more than 99% were achieved for 92.1% (neural network), 90% (support vector machine), and 43% (random forest) of these cases by applying thresholds to the resulting probability scores and excluding samples with low confidence. As independent clinical validation of the approach, we analyzed a series of 51 patients with a history of HNSC and a second lung tumor, demonstrating the correct classifications based on clinicopathological properties. In summary, our approach may facilitate the reliable diagnostic differentiation of pulmonary metastases of HNSC from primary LUSC to guide therapeutic decisions.

Computational models of cardiac electrophysiology are being investigated for improved patient selection and planning of therapies like cardiac resynchronization therapy (CRT). However, their clinical applicability is limited unless their parameters are fitted to the physiology of an individual patient. In this paper, a method that estimates spatially-varying electrical diffusivities from routine ECG data and dynamic cardiac images is presented. Contrary to current methods based on invasive electrophysiology studies or body surface potential mapping, our approach relies on widely available 12-lead ECG and motion information obtained from clinical images. First, a map of mechanical activation time is derived from a cardiac strain map. Then, regional electrical diffusivities are personalized such that the computed cardiac depolarization matches both the mechanical activation map and measured ECG features. The fit between measured and computed electrocardiography data after model personalization is evaluated on 14 dilated cardiomyopathy patients, exhibiting low mean errors in terms of the diagnostic ECG features QRS duration (0.1 ms) and electrical axis (10.6\(^{\circ }\)). The proposed regional approach outperforms global personalization when 12-lead ECG is the only electrophysiology data available. Furthermore, promising results of a preliminary CRT study on one patient demonstrate the predictive power of the personalized model.

BackgroundDespite modern pharmacotherapy and advanced implantable cardiac devices, overall prognosis and quality of life of HF patients remain poor. This is in part due to insufficient patient stratification and lack of individualized therapy planning, resulting in less effective treatments and a significant number of non-responders.Methods and resultsState-of-the-art clinical phenotyping was acquired, including magnetic resonance imaging (MRI) and biomarker assessment. An individualized, multi-scale model of heart function covering cardiac anatomy, electrophysiology, biomechanics and hemodynamics was estimated using a robust framework. The model was computed on n=46 HF patients, showing for the first time that advanced multi-scale models can be fitted consistently on large cohorts. Novel multi-scale parameters derived from the model of all cases were analyzed and compared against clinical parameters, cardiac imaging, lab tests and survival scores to evaluate the explicative power of the model and its potential for better patient stratification. Model validation was pursued by comparing clinical parameters that were not used in the fitting process against model parameters.ConclusionThis paper illustrates how advanced multi-scale models can complement cardiovascular imaging and how they could be applied in patient care. Based on obtained results, it becomes conceivable that, after thorough validation, such heart failure models could be applied for patient management and therapy planning in the future, as we illustrate in one patient of our cohort who received CRT-D implantation.