Improved Mass Calibration in MALDI MSI Using Neural Network-Based Recalibration.

Matrix-assisted laser desorption/ionization mass spectrometry imaging (MALDI MSI) is a powerful imaging method for generating molecular maps of biological samples and has numerous applications in biomedical research. A key challenge in MALDI MSI is to reliably map observed mass peaks to theoretical masses, which can be difficult due to mass shifts that occur during the measurement process. In this paper, we propose MassShiftNet, a novel self-supervised learning framework for mass recalibration. We train a neural network on a data dependent and specifically augmented training data set to directly estimate and correct the mass shift in the observed spectra. In our evaluation, we show that this method is both able to reduce the absolute mass error and to increase the relative mass alignment between peptide MSI spectra acquired from FFPE-fixated tissue using a MALDI time-of-flight (TOF) instrument.

Conditional Invertible Neural Networks for Medical Imaging.

Over recent years, deep learning methods have become an increasingly popular choice for solving tasks from the field of inverse problems. Many of these new data-driven methods have produced impressive results, although most only give point estimates for the reconstruction. However, especially in the analysis of ill-posed inverse problems, the study of uncertainties is essential. In our work, we apply generative flow-based models based on invertible neural networks to two challenging medical imaging tasks, i.e., low-dose computed tomography and accelerated medical resonance imaging. We test different architectures of invertible neural networks and provide extensive ablation studies. In most applications, a standard Gaussian is used as the base distribution for a flow-based model. Our results show that the choice of a radial distribution can improve the quality of reconstructions.

Quantitative Comparison of Deep Learning-Based Image Reconstruction Methods for Low-Dose and Sparse-Angle CT Applications.

The reconstruction of computed tomography (CT) images is an active area of research. Following the rise of deep learning methods, many data-driven models have been proposed in recent years. In this work, we present the results of a data challenge that we organized, bringing together algorithm experts from different institutes to jointly work on quantitative evaluation of several data-driven methods on two large, public datasets during a ten day sprint. We focus on two applications of CT, namely, low-dose CT and sparse-angle CT. This enables us to fairly compare different methods using standardized settings. As a general result, we observe that the deep learning-based methods are able to improve the reconstruction quality metrics in both CT applications while the top performing methods show only minor differences in terms of peak signal-to-noise ratio (PSNR) and structural similarity (SSIM). We further discuss a number of other important criteria that should be taken into account when selecting a method, such as the availability of training data, the knowledge of the physical measurement model and the reconstruction speed.

Stronger shared taste for natural aesthetic domains than for artifacts of human culture.

Individuals can be aesthetically engaged by a diverse array of visual experiences (paintings, mountain vistas, etc.), yet the processes that support this fundamental mode of interaction with the world are poorly understood. We tested whether there are systematic differences in the degree of shared taste across visual aesthetic domains. In Experiment 1, preferences were measured for five different visual aesthetic domains using a between-subjects design. The degree of agreement amongst participants differed by domain, with preferences for images of faces and landscapes containing a high proportion of shared taste, while preferences for images of exterior architecture, interior architecture and artworks reflected strong individual differences. Experiment 2 used a more powerful within-subjects design to compare the two most well matched domains-natural landscapes and exterior architecture. Agreement across individuals was significantly higher for natural landscapes than exterior architecture, with no differences in reliability. These results show that the degree of shared versus individual aesthetic preference differs systematically across visual domains, even for photographic images of real-world content. The findings suggest that the distinction between naturally occurring domains (e.g. faces and landscape) versus artifacts of human culture (e.g. architecture and artwork) is a general organizational principle governing the presence of shared aesthetic taste. We suggest that the behavioral relevance of naturally occurring domains results in information processing, and hence aesthetic experience, that is highly conserved across individuals; artifacts of human culture, which lack uniform behavioral relevance for most individuals, require the use of more individual aesthetic sensibilities that reflect varying experiences and different sources of information.