Kerstin Ritter

Di Wang, Nicolas Honnorat, Peter T. Fox et al.

Deep neural networks currently provide the most advanced and accurate machine learning models to distinguish between structural MRI scans of subjects with Alzheimer's disease and healthy controls. Unfortunately, the subtle brain alterations captured by these models are difficult to interpret because of the complexity of these multi-layer and non-linear models. Several heatmap methods have been proposed to address this issue and analyze the imaging patterns extracted from the deep neural networks, but no quantitative comparison between these methods has been carried out so far. In this work, we explore these questions by deriving heatmaps from Convolutional Neural Networks (CNN) trained using T1 MRI scans of the ADNI data set, and by comparing these heatmaps with brain maps corresponding to Support Vector Machines (SVM) coefficients. Three prominent heatmap methods are studied: Layer-wise Relevance Propagation (LRP), Integrated Gradients (IG), and Guided Grad-CAM (GGC). Contrary to prior studies where the quality of heatmaps was visually or qualitatively assessed, we obtained precise quantitative measures by computing overlap with a ground-truth map from a large meta-analysis that combined 77 voxel-based morphometry (VBM) studies independently from ADNI. Our results indicate that all three heatmap methods were able to capture brain regions covering the meta-analysis map and achieved better results than SVM coefficients. Among them, IG produced the heatmaps with the best overlap with the independent meta-analysis.

Erik Großkopf, Soumya Snigdha Kundu, Hendrik Möller et al.

The Panoptic Quality (PQ) metric is the standard for jointly evaluating instance and semantic segmentation. However, its original definition relies on a One-to-One matching between predicted and ground truth segments, which is only straightforward when the IoU threshold exceeds 0.5. Below 0.5, multiple matching strategies emerge in a poorly explored problem space. We systematically elucidate this space by recasting segment matching as a constrained bipartite assignment problem. Independently bounding the prediction- and ground-truth-side degrees yields four matching strategies: One-to-One, Many-to-One, One-to-Many, and Many-to-Many. We show that the first three are well-defined within the PQ framework, while Many-to-Many falls outside it. These strategies become relevant when instances are fragmented, adjacent objects are difficult to delineate, or annotations are noisy. Central to our framework is a vertex-based accounting of TP, FN, and FP, anchored to ground truth and predicted segments rather than to matching edges. We further show that the framework extends naturally to part-aware panoptic segmentation, and we explore part-aware evaluation on biomedical data. Across configurable case studies we report how different combinations of thresholds and matching strategies behave in practice. We release a unified open-source package built on Panoptica. It exposes Voronoi-based region-wise analysis, part-aware evaluation, and Area Under Threshold Curve computations as configurable options.

Fabian Eitel, Marc-André Schulz, Moritz Seiler et al.

By promising more accurate diagnostics and individual treatment recommendations, deep neural networks and in particular convolutional neural networks have advanced to a powerful tool in medical imaging. Here, we first give an introduction into methodological key concepts and resulting methodological promises including representation and transfer learning, as well as modelling domain-specific priors. After reviewing recent applications within neuroimaging-based psychiatric research, such as the diagnosis of psychiatric diseases, delineation of disease subtypes, normative modeling, and the development of neuroimaging biomarkers, we discuss current challenges. This includes for example the difficulty of training models on small, heterogeneous and biased data sets, the lack of validity of clinical labels, algorithmic bias, and the influence of confounding variables.

Marta Oliveira, Rick Wilming, Benedict Clark et al.

Convolutional Neural Networks (CNNs) are frequently and successfully used in medical prediction tasks. They are often used in combination with transfer learning, leading to improved performance when training data for the task are scarce. The resulting models are highly complex and typically do not provide any insight into their predictive mechanisms, motivating the field of "explainable" artificial intelligence (XAI). However, previous studies have rarely quantitatively evaluated the "explanation performance" of XAI methods against ground-truth data, and transfer learning and its influence on objective measures of explanation performance has not been investigated. Here, we propose a benchmark dataset that allows for quantifying explanation performance in a realistic magnetic resonance imaging (MRI) classification task. We employ this benchmark to understand the influence of transfer learning on the quality of explanations. Experimental results show that popular XAI methods applied to the same underlying model differ vastly in performance, even when considering only correctly classified examples. We further observe that explanation performance strongly depends on the task used for pre-training and the number of CNN layers pre-trained. These results hold after correcting for a substantial correlation between explanation and classification performance.

Fabian Eitel, Anna Melkonyan, Kerstin Ritter

A major prerequisite for the application of machine learning models in clinical decision making is trust and interpretability. Current explainability studies in the neuroimaging community have mostly focused on explaining individual decisions of trained models, e.g. obtained by a convolutional neural network (CNN). Using attribution methods such as layer-wise relevance propagation or SHAP heatmaps can be created that highlight which regions of an input are more relevant for the decision than others. While this allows the detection of potential data set biases and can be used as a guide for a human expert, it does not allow an understanding of the underlying principles the model has learned. In this study, we instead show, to the best of our knowledge, for the first time results using feature visualization of neuroimaging CNNs. Particularly, we have trained CNNs for different tasks including sex classification and artificial lesion classification based on structural magnetic resonance imaging (MRI) data. We have then iteratively generated images that maximally activate specific neurons, in order to visualize the patterns they respond to. To improve the visualizations we compared several regularization strategies. The resulting images reveal the learned concepts of the artificial lesions, including their shapes, but remain hard to interpret for abstract features in the sex classification task.

Roshan Prakash Rane, JiHoon Kim, Arjun Umesha et al.

Deep Learning (DL) models have gained popularity in neuroimaging studies for predicting psychological behaviors, cognitive traits, and brain pathologies. However, these models can be biased by confounders such as age, sex, or imaging artifacts from the acquisition process. To address this, we introduce 'DeepRepViz', a two-part framework designed to identify confounders in DL model predictions. The first component is a visualization tool that can be used to qualitatively examine the final latent representation of the DL model. The second component is a metric called 'Con-score' that quantifies the confounder risk associated with a variable, using the final latent representation of the DL model. We demonstrate the effectiveness of the Con-score using a simple simulated setup by iteratively altering the strength of a simulated confounder and observing the corresponding change in the Con-score. Next, we validate the DeepRepViz framework on a large-scale neuroimaging dataset (n=12000) by performing three MRI-phenotype prediction tasks that include (a) predicting chronic alcohol users, (b) classifying participant sex, and (c) predicting performance speed on a cognitive task called 'trail making'. DeepRepViz identifies sex as a significant confounder in the DL model predicting chronic alcohol users (Con-score=0.35) and age as a confounder in the model predicting cognitive task performance (Con-score=0.3). In conclusion, the DeepRepViz framework provides a systematic approach to test for potential confounders such as age, sex, and imaging artifacts and improves the transparency of DL models for neuroimaging studies.

Sam Gijsen, Kerstin Ritter

Multimodal language modeling has enabled breakthroughs for representation learning, yet remains unexplored in the realm of functional brain data for clinical phenotyping. This paper pioneers EEG-language models (ELMs) trained on clinical reports and 15000 EEGs. We propose to combine multimodal alignment in this novel domain with timeseries cropping and text segmentation, enabling an extension based on multiple instance learning to alleviate misalignment between irrelevant EEG or text segments. Our multimodal models significantly improve over EEG-only models across four clinical evaluations and for the first time enable zero-shot classification as well as retrieval of both neural signals and reports. In sum, these results highlight the potential of ELMs, representing significant progress for clinical applications.

Marc-Andre Schulz, Kerstin Ritter

On biomedical tabular data, flexible models such as deep networks, gradient-boosted trees, and kernel methods are repeatedly matched or beaten by linear and logistic regression given the same features. The usual reaction is to treat this as a model-side shortfall, to be fixed with more data, a better architecture, or tuning, on the assumption that the nonlinear structure is there and the model has failed to capture it. We argue that these fixes cannot help when the binding limit is the measurement rather than the model, as it frequently is in biomedicine. Additive noise blurs the population-optimal predictor, and because blurring removes a function's fine, rapidly varying detail before its broad shape, it erases nonlinear structure faster than linear structure. A degree-$k$ interaction is attenuated by the $k$-th power of feature reliability, while the linear part is attenuated only once. At the reliabilities typical of biomedical measurement, the nonlinear advantage can vanish even when the underlying biology is strongly nonlinear, and what the noise removes cannot be recovered by a larger cohort or a more flexible model, only by better measurement. The nonlinearity is hidden, not absent, and a tie between linear and flexible models is not by itself a verdict on the biology. These pieces are classical, drawn from measurement-error statistics, psychometrics, and Gaussian analysis, and we assemble them into an exact excess-risk identity. Measurement reliability is one of three conditions, alongside sample size and feature representation, that must align for a flexible model to help, and together they leave only a narrow window that most biomedical tasks fall outside. Across 140 UK Biobank tasks, the gap between flexible and linear models, where it exists, carries the predicted noise signature, and the three conditions can be separated by intervention but not by a benchmark alone.

Céline Budding, Fabian Eitel, Kerstin Ritter et al.

Over the last years, many 'explainable artificial intelligence' (xAI) approaches have been developed, but these have not always been objectively evaluated. To evaluate the quality of heatmaps generated by various saliency methods, we developed a framework to generate artificial data with synthetic lesions and a known ground truth map. Using this framework, we evaluated two data sets with different backgrounds, Perlin noise and 2D brain MRI slices, and found that the heatmaps vary strongly between saliency methods and backgrounds. We strongly encourage further evaluation of saliency maps and xAI methods using this framework before applying these in clinical or other safety-critical settings.

Fabian Eitel, Jan Philipp Albrecht, Martin Weygandt et al.

Neuroimaging data, e.g. obtained from magnetic resonance imaging (MRI), is comparably homogeneous due to (1) the uniform structure of the brain and (2) additional efforts to spatially normalize the data to a standard template using linear and non-linear transformations. Convolutional neural networks (CNNs), in contrast, have been specifically designed for highly heterogeneous data, such as natural images, by sliding convolutional filters over different positions in an image. Here, we suggest a new CNN architecture that combines the idea of hierarchical abstraction in neural networks with a prior on the spatial homogeneity of neuroimaging data: Whereas early layers are trained globally using standard convolutional layers, we introduce for higher, more abstract layers patch individual filters (PIF). By learning filters in individual image regions (patches) without sharing weights, PIF layers can learn abstract features faster and with fewer samples. We thoroughly evaluated PIF layers for three different tasks and data sets, namely sex classification on UK Biobank data, Alzheimer's disease detection on ADNI data and multiple sclerosis detection on private hospital data. We demonstrate that CNNs using PIF layers result in higher accuracies, especially in low sample size settings, and need fewer training epochs for convergence. To the best of our knowledge, this is the first study which introduces a prior on brain MRI for CNN learning.

Fabian Eitel, Jan Philipp Albrecht, Friedemann Paul et al.

Neuroimaging studies based on magnetic resonance imaging (MRI) typically employ rigorous forms of preprocessing. Images are spatially normalized to a standard template using linear and non-linear transformations. Thus, one can assume that a patch at location (x, y, height, width) contains the same brain region across the entire data set. Most analyses applied on brain MRI using convolutional neural networks (CNNs) ignore this distinction from natural images. Here, we suggest a new layer type called patch individual filter (PIF) layer, which trains higher-level filters locally as we assume that more abstract features are locally specific after spatial normalization. We evaluate PIF layers on three different tasks, namely sex classification as well as either Alzheimer's disease (AD) or multiple sclerosis (MS) detection. We demonstrate that CNNs using PIF layers outperform their counterparts in several, especially low sample size settings.

Fabian Eitel, Kerstin Ritter

Attribution methods are an easy to use tool for investigating and validating machine learning models. Multiple methods have been suggested in the literature and it is not yet clear which method is most suitable for a given task. In this study, we tested the robustness of four attribution methods, namely gradient*input, guided backpropagation, layer-wise relevance propagation and occlusion, for the task of Alzheimer's disease classification. We have repeatedly trained a convolutional neural network (CNN) with identical training settings in order to separate structural MRI data of patients with Alzheimer's disease and healthy controls. Afterwards, we produced attribution maps for each subject in the test data and quantitatively compared them across models and attribution methods. We show that visual comparison is not sufficient and that some widely used attribution methods produce highly inconsistent outcomes.

Fabian Eitel, Emily Soehler, Judith Bellmann-Strobl et al.

Machine learning-based imaging diagnostics has recently reached or even superseded the level of clinical experts in several clinical domains. However, classification decisions of a trained machine learning system are typically non-transparent, a major hindrance for clinical integration, error tracking or knowledge discovery. In this study, we present a transparent deep learning framework relying on convolutional neural networks (CNNs) and layer-wise relevance propagation (LRP) for diagnosing multiple sclerosis (MS). MS is commonly diagnosed utilizing a combination of clinical presentation and conventional magnetic resonance imaging (MRI), specifically the occurrence and presentation of white matter lesions in T2-weighted images. We hypothesized that using LRP in a naive predictive model would enable us to uncover relevant image features that a trained CNN uses for decision-making. Since imaging markers in MS are well-established this would enable us to validate the respective CNN model. First, we pre-trained a CNN on MRI data from the Alzheimer's Disease Neuroimaging Initiative (n = 921), afterwards specializing the CNN to discriminate between MS patients and healthy controls (n = 147). Using LRP, we then produced a heatmap for each subject in the holdout set depicting the voxel-wise relevance for a particular classification decision. The resulting CNN model resulted in a balanced accuracy of 87.04% and an area under the curve of 96.08% in a receiver operating characteristic curve. The subsequent LRP visualization revealed that the CNN model focuses indeed on individual lesions, but also incorporates additional information such as lesion location, non-lesional white matter or gray matter areas such as the thalamus, which are established conventional and advanced MRI markers in MS. We conclude that LRP and the proposed framework have the capability to make diagnostic decisions of...

Johannes Rieke, Fabian Eitel, Martin Weygandt et al.

Visualizing and interpreting convolutional neural networks (CNNs) is an important task to increase trust in automatic medical decision making systems. In this study, we train a 3D CNN to detect Alzheimer's disease based on structural MRI scans of the brain. Then, we apply four different gradient-based and occlusion-based visualization methods that explain the network's classification decisions by highlighting relevant areas in the input image. We compare the methods qualitatively and quantitatively. We find that all four methods focus on brain regions known to be involved in Alzheimer's disease, such as inferior and middle temporal gyrus. While the occlusion-based methods focus more on specific regions, the gradient-based methods pick up distributed relevance patterns. Additionally, we find that the distribution of relevance varies across patients, with some having a stronger focus on the temporal lobe, whereas for others more cortical areas are relevant. In summary, we show that applying different visualization methods is important to understand the decisions of a CNN, a step that is crucial to increase clinical impact and trust in computer-based decision support systems.