Importance The entry of artificial intelligence into medicine is pending. Several methods have been used for the predictions of structured neuroimaging data, yet nobody compared them in this context. Objective Multi-class prediction is key for building computational aid systems for differential diagnosis. We compared support vector machine, random forest, gradient boosting, and deep feed-forward neural networks for the classification of different neurodegenerative syndromes based on structural magnetic resonance imaging. Design, setting, and participants Atlas-based volumetry was performed on multi-centric T1-weighted MRI data from 940 subjects, i.e., 124 healthy controls and 816 patients with ten different neurodegenerative diseases, leading to a multi-diagnostic multi-class classification task with eleven different classes. Interventions N.A. Main outcomes and measures Cohen’s kappa, accuracy, and F1-score to assess model performance. Results Overall, the neural network produced both the best performance measures and the most robust results. The smaller classes however were better classified by either the ensemble learning methods or the support vector machine, while performance measures for small classes were comparatively low, as expected. Diseases with regionally specific and pronounced atrophy patterns were generally better classified than diseases with widespread and rather weak atrophy. Conclusions and relevance Our study furthermore underlines the necessity of larger data sets but also calls for a careful consideration of different machine learning methods that can handle the type of data and the classification task best.

We present a reinforcement learning approach for detecting objects within an image. Our approach performs a step-wise deformation of a bounding box with the goal of tightly framing the object. It uses a hierarchical tree-like representation of predefined region candidates, which the agent can zoom in on. This reduces the number of region candidates that must be evaluated so that the agent can afford to compute new feature maps before each step to enhance detection quality. We compare an approach that is based purely on zoom actions with one that is extended by a second refinement stage to fine-tune the bounding box after each zoom step. We also improve the fitting ability by allowing for different aspect ratios of the bounding box. Finally, we propose different reward functions to lead to a better guidance of the agent while following its search trajectories. Experiments indicate that each of these extensions leads to more correct detections. The best performing approach comprises a zoom stage and a refinement stage, uses aspect-ratio modifying actions and is trained using a combination of three different reward metrics.

Machine learning has considerably improved medical image analysis in the past years. Although datadriven approaches are intrinsically adaptive and thus, generic, they often do not perform the same way on data from different imaging modalities. In particular computed tomography (CT) data poses many challenges to medical image segmentation based on convolutional neural networks (CNNs), mostly due to the broad dynamic range of intensities and the varying number of recorded slices of CT volumes. In this paper, we address these issues with a framework that adds domain-specific data preprocessing and augmentation to state-of-the-art CNN architectures. Our major focus is to stabilise the prediction performance over samples as a mandatory requirement for use in automated and semiautomated workflows in the clinical environment. To validate the architecture-independent effects of our approach we compare a neural architecture based on dilated convolutions for parallel multi-scale processing (a modified Mixed-Scale Dense Network: MS-D Net) to traditional scaling operations (a modified U-Net). Finally, we show that an ensemble model combines the strengths across different individual methods. Our framework is simple to implement into existing deep learning pipelines for CT analysis. It performs well on a range of tasks such as liver and kidney segmentation, without significant differences in prediction performance on strongly differing volume sizes and varying slice thickness. Thus our framework is an essential step towards performing robust segmentation of unknown real-world samples. The geometry of tumours, as described by, e.g. its size, shape or location, is a central clinical feature. Changes in these geometric characteristics are essential indicators of disease progression and can be used to measure treatment effects. An automated, quantitative assessment of these aspects and their changes from radiological images would yield an efficient and objective tool for radiologists to monitor the course of the disease. Thus, a reliable and accurate automated segmentation method is desirable to extract spatial tumour and organ characteristics from computed tomography (CT) volumes. In recent years, convolutional neural networks (CNNs) 1 became the state of the art method for image segmentation, as well as many other tasks in computer vision 2 , such as image classification, object detection and object tracking 3. The applications of CNNs are diverse, but the general data handling or preprocessing is often very similar in each case since the feature extraction is performed internally by the CNN itself. Improvements in the application of CNNs for medical image processing often address changes in the neural network architecture, the training algorithm or the use case 4,5. At the same time, most authors tend to ignore the data handling itself, treating medical images such as CT volumes the same way as grayscale or RGB images with additional dimensions. However, this approach neglects prior information about the specific physical proce...