Active learning improves annotation efficiency by selecting the most informative samples for annotation and model training. While most prior work has focused on selecting informative images for classification tasks, we investigate the more challenging setting of dense prediction, where annotations are more costly and time-intensive, especially in medical imaging. Region-level annotation has been shown to be more efficient than image-level annotation for these tasks. However, existing methods for representative annotation region selection suffer from high computational and memory costs, irrelevant region choices, and heavy reliance on uncertainty sampling. We propose decomposition sampling (DECOMP), a new active learning sampling strategy that addresses these limitations. It enhances annotation diversity by decomposing images into class-specific components using pseudo-labels and sampling regions from each class. Class-wise predictive confidence further guides the sampling process, ensuring that difficult classes receive additional annotations. Across ROI classification, 2-D segmentation, and 3-D segmentation, DECOMP consistently surpasses baseline methods by better sampling minority-class regions and boosting performance on these challenging classes. Code is in https://github.com/JingnaQiu/DECOMP.git.

As deep learning models continue to scale in complexity and data size, reducing storage and training costs has become increasingly important. Dataset distillation addresses this challenge by synthesizing a small set of synthetic samples that effectively substitute for the original dataset in downstream tasks. Existing approaches typically rely on matching gradients or features either in pixel space or in the latent space of a pretrained generative model. We propose a novel stochastic distillation method that models the joint distribution of latent features using a low-rank multivariate normal distribution, parameterized by a lightweight neural network. This formulation captures spatial correlations in the feature space, which are then projected into class probability space to generate more diverse and informative predictions. The proposed module integrates seamlessly with existing distillation pipelines. Our method achieves state-of-the-art cross-architecture results, improving test accuracy by up to 7.47% in gradient matching and 35.71% in distribution matching over baselines. • Introduce SLFD, a framework that distills data with stochastic latent features. • Model spatial correlations using a low-rank multivariate distribution. • Achieve robust performance on high-resolution ImageNet-1K subsets. • Demonstrate applicability to medical imaging with strong results.

scores of up to 80.1% and 77.19% on the LSCC and LUAD datasets, respectively, and with 77.09% on our meningioma dataset. Based on our findings, we formulated a risk assessment scheme to estimate the risk to the patient's privacy prior to publication.

Object detection is one of the most common tasks in histopathological image analysis and generalization is a key requirement for the clinical applicability of deep object detection models. However, traditional evaluation metrics often fail to provide insights into why models fail on certain test cases, especially in the presence of domain shifts. In this work, we propose a novel quantitative method for assessing the discriminative power of a model's latent space. Our approach, applicable to all object detection models with known local correspondences such as the popular RetinaNet, FCOS, or YOLO approaches, allows tracing discrimination across layers and coordinates. We motivate, adapt, and evaluate two suitable metrics, the generalized discrimination value and the Hellinger distance, and incorporate them into our approach. Through empirical validation on real-world histopathology datasets, we demonstrate the effectiveness of our method in capturing model discrimination properties and providing insights for architectural optimization. This work contributes to bridging the gap between model performance evaluation and understanding the underlying mechanisms influencing model behavior.

Magnetic resonance imaging (MRI) enables 3-D imaging of anatomical structures. However, the acquisition of MR volumes with high spatial resolution leads to long scan times. To this end, we propose volumetric super-resolution forests (VSRF) to enhance MRI resolution retrospectively. Our method learns a locally linear mapping between low-resolution and high-resolution volumetric image patches by employing random forest regression. We customize features suitable for volumetric MRI to train the random forest and propose a median tree ensemble for robust regression. VSRF out-performs state-of-the-art example-based super-resolution in terms of image quality and efficiency for model training and inference on different MRI datasets. It is also superior to unsupervised methods with just a handful or even a single volume to assemble training data.

Late-gadolinium enhanced magnetic resonance imaging (LGE-MRI) is the clinical gold standard to visualize myocardial scarring. The gadolinium based contrast agent accumulates in the damaged cells and leads to various enhancements in the LGE-MRI scan. The quantification of the scar tissue is very important for diagnosis, treatment planning, and guidance during the procedure. In clinical routine, the scar is often segmented manually. However, manual segmentation is prone to inter- and intra-observer variability and very time consuming. In this work a new texture based scar quantification is proposed. For texture characterization, segmentation based fractal analysis is used. First, the image is decomposed into a set of binary images by applying a two-threshold binary decomposition. Second, a set of features are extracted for each of the binary images, namely the fractal dimension, the mean gray value, and the size of the binary object. In addition, the local and global intensity of each patch is added to the feature vector. In the next step, the features are classified using a random forest classifier. The scar quantification is evaluated on 30 clinical LGE-MRI data sets. In addition, the results are compared to the x-fold standard deviation approach and the full-width-at-half-max method, which are implemented in a fully automatic manner. The proposed scar quantification achieved a mean Dice coefficient of 0.64±0.17 and outperforms the x-fold standard deviation approach.

In this paper, we present a new deep learning framework for 3-D tomographic reconstruction. To this end, we map filtered back-projection-type algorithms to neural networks. However, the back-projection cannot be implemented as a fully connected layer due to its memory requirements. To overcome this problem, we propose a new type of cone-beam back-projection layer, efficiently calculating the forward pass. We derive this layer's backward pass as a projection operation. Unlike most deep learning approaches for reconstruction, our new layer permits joint optimization of correction steps in volume and projection domain. Evaluation is performed numerically on a public data set in a limited angle setting showing a consistent improvement over analytical algorithms while keeping the same computational test-time complexity by design. In the region of interest, the peak signal-to-noise ratio has increased by 23%. In addition, we show that the learned algorithm can be interpreted using known concepts from cone beam reconstruction: the network is able to automatically learn strategies such as compensation weights and apodization windows.