Ludwig Lausser's research while affiliated with Ulm University and other places

204 Background: Anti-vascular endothelial growth factor (VEGF) monoclonal antibodies (mAbs) are widely used for tumor treatment, including metastatic colorectal cancer (mCRC). So far, there are no biomarkers that reliably predict resistance to anti-VEGF mAbs like bevacizumab. A biomarker-guided strategy for early and accurate assessment of resistance could avoid the use of non-effective treatment and improve patient outcomes. We hypothesized that repeated analysis of multiple cytokines and angiogenic growth factors (CAFs) before and during treatment using machine learning could provide an accurate and earlier, i.e., 100 days before conventional radiologic staging, prediction of resistance to first-line mCRC treatment with FOLFOX plus bevacizumab. Methods: 15 German and Austrian centers prospectively recruited 154 mCRC patients receiving FOLFOX plus bevacizumab as first-line treatment. Plasma samples were collected every two weeks until radiologic progression (RECIST 1.1) as determined by CT scans performed every 2 months. 102 pre-selected CAFs were centrally analyzed using a cytokine multiplex assay (Luminex, Myriad RBM). Results: Using random forest machine learning, we developed a predictive model that discriminated between the situations of ”no progress within 100 days before radiological progress” and ”progress within 100 days before radiological progress”. Into this we incorporated a combination of ten out of the 102 CAF markers, which fulfilled this task with 81% accuracy, 72% sensitivity, and 88% specificity. Conclusions: Using artificial intelligence we identified a CAF marker combination that indicates treatment resistance to FOLFOX plus bevacizumab in patients with mCRC within 100 days prior to radiologic progress. Further studies are required to show its clinical value. Clinical trial information: NCT02331927 .

Pancreatic neuroendocrine tumors (PanNETs) are a rare tumor entity with largely unpredictable progression and increasing incidence in developed countries. Molecular pathways involved in PanNETs development are still not elucidated, and specific biomarkers are missing. Moreover, the heterogeneity of PanNETs makes their treatment challenging and most approved targeted therapeutic options for PanNETs lack objective responses. Here, we applied a systems biology approach integrating dynamic modeling strategies, foreign classifier tailored approaches, and patient expression profiles to predict PanNETs progression as well as resistance mechanisms to clinically approved treatments such as the mammalian target of rapamycin complex 1 (mTORC1) inhibitors. We set up a model able to represent frequently reported PanNETs drivers in patient cohorts, such as Menin-1 (MEN1), Death domain associated protein (DAXX), Tuberous Sclerosis (TSC), as well as wild-type tumors. Model-based simulations suggested drivers of cancer progression as both first and second hits after MEN1 loss. In addition, we could predict the benefit of mTORC1 inhibitors on differentially mutated cohorts and hypothesize resistance mechanisms. Our approach sheds light on a more personalized prediction and treatment of PanNET mutant phenotypes.

Computer vision classification tasks rely on the availability of ground truth labels. Especially in medical imaging, these are typically given by experts and can be of differing quality. To reduce the expert bias influence on labels, commonly blinded multi-expert consensus labels are used as ground truth in machine learning. In this work, we approach the question of how good a multiexpert consensus can be for the example of mitotic figure (MF) identification, which is a relevant task in tumor malignancy assessment. For this, we provide an exhaustive evaluation of all possible majority ensembles of 23 pathologists who independently assessed MFs based on a preselected region of interest. We compared the ensemble against a immunohistochemistry-based ground truth. We found that there were upper bounds to the recognition of MFs by the experts, which were, in our dataset, an accuracy, sensitivity and specificity of 88%, 82%, and 100%, respectively. An analysis of our results revealed cells in prophase and blurry cells to be amongst the most challenging to recognize.

In recent years, several deep learning approaches have been successfully applied in the field of medical image analysis. More specifically, different deep neural network architectures have been proposed and assessed for the detection of various pathologies based on chest X-ray images. While the performed assessments have shown very promising results, most of them consist in training and evaluating the performance of the proposed approaches on a single data set. However, the generalization of such models is quite limited in a cross-domain setting, since a significant performance degradation can be observed when these models are evaluated on data sets stemming from different medical centers or recorded under different protocols. The performance degradation is mostly caused by the domain shift between the training set and the evaluation set. To alleviate this problem, different unsupervised domain adaptation approaches are proposed and evaluated in the current work, for the detection of cardiomegaly based on chest X-ray images, in a cross-domain setting. The proposed approaches generate domain invariant feature representations by adapting the parameters of a model optimized on a large set of labeled samples, to a set of unlabeled images stemming from a different data set. The performed evaluation points to the effectiveness of the proposed approaches, since the adapted models outperform optimized models which are directly applied to the evaluation sets without any form of domain adaptation.

Sparse and robust classification models have the potential for revealing common predictive patterns that not only allow for categorizing objects into classes but also for generating mechanistic hypotheses. Identifying a small and informative subset of features is their main ingredient. However, the exponential search space of feature subsets and the heuristic nature of selection algorithms limit the coverage of these analyses, even for low-dimensional datasets. We present methods for reducing the computational complexity of feature selection criteria allowing for higher efficiency and coverage of screenings. We achieve this by reducing the preparation costs of high-dimensional subsets O(nm2)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {O}}({n}m^2)$$\end{document} to those of one-dimensional ones O(m2)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {O}}(m^2)$$\end{document}. Our methods are based on a tight interaction between a parallelizable cross-validation traversal strategy and distance-based classification algorithms and can be used with any product distance or kernel. We evaluate the traversal strategy exemplarily in exhaustive feature subset selection experiments (perfect coverage). Its runtime, fitness landscape, and predictive performance are analyzed on publicly available datasets. Even in low-dimensional settings, we achieve approximately a 15-fold increase in exhaustively generating distance matrices for feature combinations bringing a new level of evaluations into reach.

We here respond to the points raised in a recent letter to the editor on the feasibility of dynamical analyses in Boolean networks, referring to our manuscript ”Capturing dynamic relevance in Boolean networks using graph theoretical measures”.

Ordinal classification (OC) is a sub-discipline of multi-class classification (i.e., including at least three classes), in which the classes constitute an ordinal structure. Applications of ordinal classification can be found, for instance, in the medical field, e.g., with the class labels order, early stage-intermediate stage-final stage, corresponding to the task of classifying different stages of a certain disease. While the field of OC was continuously enhanced, e.g., by designing and adapting appropriate classification models as well as performance metrics, there is still a lack of a common mathematical definition for OC tasks. More precisely, in general, a classification task is defined as an OC task, solely based on the corresponding class label names. However, an ordinal class structure that is identified based on the class labels is not necessarily reflected in the corresponding feature space. In contrast, naturally any kind of multi-class classification task can consist of a set of arbitrary class labels that form an ordinal structure which can be observed in the current feature space. Based on this simple observation, in this work, we present our generalised approach towards an intuitive working definition for OC tasks, which is based on the corresponding feature space and allows a classifier-independent detection of ordinal class structures. To this end, we introduce and discuss novel, OC-specific theoretical concepts. Moreover, we validate our proposed working definition in combination with a set of traditionally ordinal and traditionally non-ordinal data sets, and provide the results of the corresponding detection algorithm. Additionally, we motivate our theoretical concepts, based on an illustrative evaluation of one of the oldest and most popular machine learning data sets, i.e., on the traditionally non-ordinal Fisher’s Iris data set.

Introduction: The behavioral variant of frontotemporal dementia (bvFTD) is a rare neurodegenerative disease. Reliable predictors of disease progression have not been sufficiently identified. We investigated multivariate magnetic resonance imaging (MRI) biomarker profiles for their predictive value of individual decline. Methods: One hundred five bvFTD patients were recruited from the German frontotemporal lobar degeneration (FTLD) consortium study. After defining two groups ("fast progressors" vs. "slow progressors"), we investigated the predictive value of MR brain volumes for disease progression rates performing exhaustive screenings with multivariate classification models. Results: We identified areas that predict disease progression rate within 1 year. Prediction measures revealed an overall accuracy of 80% across our 50 top classification models. Especially the pallidum, middle temporal gyrus, inferior frontal gyrus, cingulate gyrus, middle orbitofrontal gyrus, and insula occurred in these models. Discussion: Based on the revealed marker combinations an individual prognosis seems to be feasible. This might be used in clinical studies on an individualized progression model.

Alzheimer’s disease (AD), the most prevalent form of dementia, affects globally more than 30 million people suffering from cognitive deficits and neuropsychiatric symptoms. Substantial evidence for the involvement of mitochondrial dysfunction in the development and/or progression of AD has been shown in addition to the pathological hallmarks amyloid beta (Aβ) and tau. Still, the selective vulnerability and associated selective mitochondrial dysfunction cannot even be resolved to date. We aimed at optically quantifying mitochondrial function on a single-cell level in primary hippocampal neuron models of AD, unraveling differential involvement of cell and mitochondrial populations in amyloid precursor protein (APP)-associated mitochondrial dysfunction. NADH lifetime imaging is a highly sensitive marker-free method with high spatial resolution. However, deciphering cellular bioenergetics of complex cells like primary neurons has still not succeeded yet. To achieve this, we combined highly sensitive NADH lifetime imaging with respiratory inhibitor treatment, allowing characterization of mitochondrial function down to even the subcellular level in primary neurons. Measuring NADH lifetime of the same neuron before and after respiratory treatment reveals the metabolic delta, which can be taken as a surrogate for cellular redox capacity. Correlating NADH lifetime delta with overexpression strength of Aβ-related proteins on the single-cell level, we could verify the important role of intracellular Aβ-mediated mitochondrial toxicity. Subcellularly, we could demonstrate a higher respiration in neuronal somata in general than dendrites, but a similar impairment of somatic and dendritic mitochondria in our AD models. This illustrates the power of NADH lifetime imaging in revealing mitochondrial function on a single and even subcellular level and its potential to shed light into bioenergetic alterations in neuropsychiatric diseases and beyond.