Spatial transcriptomic technologies are transforming our ability to profile gene expression, by mapping it to the native architecture of the tissues. However, the inherent complexity and multimodality of the problem demand robust analytical computational tools. In this study, we benchmarked seven state-of-the art clustering algorithms that cover different methodological approaches-conventional statistics, Machine Learning, and Deep Learning-based on their capacity to integrate gene expression, histology, and spatial proximity to identify spatial domains in spatial transcriptomic data. Our comparative analysis implemented on a wellannotated human brain reference dataset shows that algorithms that integrate spatial adjacency (GraphST and BayesSpace) significantly outperform the others, with GraphST achieving the best performance. As a proof of principle, we applied GraphST to a highgrade serous ovarian cancer sample. The model revealed transcriptionally distinct subregions within tumor and stromal compartments not distinguishable by histological assessment alone. Furthermore, by integrating cell-type deconvolution with spatial T-cell receptor (TCR) clonotyping, we discovered heterogeneous immune niches, characterized by domain-specific clonal expansion. Tumor cores exhibited highly diverse TCR repertoires, while stromal margins were dominated by a few expanded clones, suggesting differential immune-selection pressure. These findings highlight the potential of spatially informed Deep Learning methods to dissect the tumor microenvironment and identify actionable targets for immune-based anticancer therapies.

Multiplex immunofluorescence is a novel, high-content imaging technique that allows simultaneous in situ labeling of multiple tissue antigens. This technique is of growing relevance in the study of the tumor microenvironment, and the discovery of biomarkers of disease progression or response to immune-based therapies. Given the number of markers and the potential complexity of the spatial interactions involved, the analysis of these images requires the use of machine learning tools that rely for their training on the availability of large image datasets, extremely laborious to annotate. We present Synplex, a computer simulator of multiplexed immunofluorescence images from user-defined parameters: i. cell phenotypes, defined by the level of expression of markers and morphological parameters; ii. cellular neighborhoods based on the spatial association of cell phenotypes; and iii. interactions between cellular neighborhoods. We validate Synplex by generating synthetic tissues that accurately simulate real cancer cohorts with underlying differences in the composition of their tumor microenvironment and show proof-of-principle examples of how Synplex could be used for data augmentation when training machine learning models, and for the in silico selection of clinically relevant biomarkers. Synplex is publicly available at https://github.com/djimenezsanchez/Synplex.

Understanding the spatial interactions between the elements of the tumor microenvironment -i.e. tumor cells. fibroblasts, immune cells- and how these interactions relate to the diagnosis or prognosis of a tumor is one of the goals of computational pathology. We present NaroNet, a deep learning framework that models the multi-scale tumor microenvironment from multiplex-stained cancer tissue images and provides patient-level interpretable predictions using a seamless end-to-end learning pipeline. Trained only with multiplex-stained tissue images and their corresponding patient-level clinical labels, NaroNet unsupervisedly learns which cell phenotypes, cell neighborhoods, and neighborhood interactions have the highest influence to predict the correct label. To this end, NaroNet incorporates several novel and state-of-the-art deep learning techniques, such as patch-level contrastive learning, multi-level graph embeddings, a novel max-sum pooling operation, or a metric that quantifies the relevance that each microenvironment element has in the individual predictions. We validate NaroNet using synthetic data simulating multiplex-immunostained images where a patient label is artificially associated to the -adjustable- probabilistic incidence of different microenvironment elements. We then apply our model to two sets of images of human cancer tissues: 336 seven-color multiplex-immunostained images from 12 high-grade endometrial cancer patients; and 382 35-plex mass cytometry images from 215 breast cancer patients. In both synthetic and real datasets, NaroNet provides outstanding predictions of relevant clinical information while associating those predictions to the presence of specific microenvironment elements.

MOTIVATION: Recent advances in multiplex immunostaining and multispectral cytometry have opened the door to simultaneously visualizing an unprecedented number of biomarkers both in liquid and solid samples. Properly unmixing fluorescent emissions is a challenging task, which normally requires the characterization of the individual fluorochromes from control samples. As the number of fluorochromes increases, the cost in time and use of reagents becomes prohibitively high. Here, we present a fully unsupervised blind spectral unmixing method for the separation of fluorescent emissions in highly mixed spectral data, without the need for control samples. To this end, we extend an existing method based on non-negative Matrix Factorization, and introduce several critical improvements: initialization based on the theoretical spectra, automated selection of 'sparse' data and use of a re-initialized multilayer optimizer. RESULTS: Our algorithm is exhaustively tested using synthetic data to study its robustness against different levels of colocalization, signal to noise ratio, spectral resolution and the effect of errors in the initialization of the algorithm. Then, we compare the performance of our method to that of traditional spectral unmixing algorithms using novel multispectral flow and image cytometry systems. In all cases, we show that our blind unmixing algorithm performs robust unmixing of highly spatially and spectrally mixed data with an unprecedently low computational cost. In summary, we present the first use of a blind unmixing method in multispectral flow and image cytometry, opening the door to the widespread use of our method to efficiently pre-process multiplex immunostaining samples without the need of experimental controls. AVAILABILITY AND IMPLEMENTATION: https://github.com/djimenezsanchez/Blind_Unmixing_NMF_RI/ contains the source code and all datasets used in this manuscript. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

We present a dynamic atlas composed of neuromelanin-enhanced magnetic resonance brain images of 40 healthy subjects. The performance of this atlas is evaluated on the fully automated segmentation of two paired neuromelanin-rich brainstem healthy structures: the substantia nigra pars compacta and the locus coeruleus. We show that our dynamic atlas requires in average 60% less images and, therefore, 60% less computation time than a static multi-image atlas while achieving a similar segmentation performance. Then, we show that by applying our dynamic atlas, composed of healthy subjects, to the segmentation and neuromelanin quantification of a set of brain images of 39 Parkinson disease patients, we are able to find significant quantitative differences in the level of neuromelanin between healthy subjects and Parkinson disease patients, thus opening the door to the use of these structures as image biomarkers in future computer aided diagnosis systems for the diagnosis of Parkinson disease.

We present a 3D bioimage analysis workflow to quantitatively analyze single, actin-stained cells with filopodial protrusions of diverse structural and temporal attributes, such as number, length, thickness, level of branching, and lifetime, in time-lapse confocal microscopy image data. Our workflow makes use of convolutional neural networks trained using real as well as synthetic image data, to segment the cell volumes with highly heterogeneous fluorescence intensity levels and to detect individual filopodial protrusions, followed by a constrained nearest-neighbor tracking algorithm to obtain valuable information about the spatio-temporal evolution of individual filopodia. We validated the workflow using real and synthetic 3-D time-lapse sequences of lung adenocarcinoma cells of three morphologically distinct filopodial phenotypes and show that it achieves reliable segmentation and tracking performance, providing a robust, reproducible and less time-consuming alternative to manual analysis of the 3D+t image data.

The geometry of 3D collagen networks is a key factor that influences the behavior of live cells within extracellular matrices. This paper presents a hybrid, two-step method for fully automatic quantification of the 3D collagen network geometry at fiber resolution in confocal reflection microscopy images. First, a coarse binary mask of the entire network is obtained using steerable filtering and local Otsu thresholding. Second, individual collagen fibers are reconstructed by tracing maximum ridges in the Euclidean distance map of the binary mask. The proposed method, validated in a novel framework using 3D collagen gels with various collagen and Matrigel concentrations, reveal that Matrigel affects the collagen network geometry by decreasing the network pore size while preserving the fiber length and fiber persistence length.

MOTIVATION: Automatic tracking of cells in multidimensional time-lapse fluorescence microscopy is an important task in many biomedical applications. A novel framework for objective evaluation of cell tracking algorithms has been established under the auspices of the IEEE International Symposium on Biomedical Imaging 2013 Cell Tracking Challenge. In this article, we present the logistics, datasets, methods and results of the challenge and lay down the principles for future uses of this benchmark. RESULTS: The main contributions of the challenge include the creation of a comprehensive video dataset repository and the definition of objective measures for comparison and ranking of the algorithms. With this benchmark, six algorithms covering a variety of segmentation and tracking paradigms have been compared and ranked based on their performance on both synthetic and real datasets. Given the diversity of the datasets, we do not declare a single winner of the challenge. Instead, we present and discuss the results for each individual dataset separately. AVAILABILITY AND IMPLEMENTATION: The challenge Web site (http://www.codesolorzano.com/celltrackingchallenge) provides access to the training and competition datasets, along with the ground truth of the training videos. It also provides access to Windows and Linux executable files of the evaluation software and most of the algorithms that competed in the challenge.

We present a fast and robust approach to tracking the evolving shape of whole fluorescent cells in time-lapse series. The proposed tracking scheme involves two steps. First, coherence-enhancing diffusion filtering is applied on each frame to reduce the amount of noise and enhance flow-like structures. Second, the cell boundaries are detected by minimizing the Chan-Vese model in the fast level set-like and graph cut frameworks. To allow simultaneous tracking of multiple cells over time, both frameworks have been integrated with a topological prior exploiting the object indication function. The potential of the proposed tracking scheme and the advantages and disadvantages of both frameworks are demonstrated on 2-D and 3-D time-lapse series of rat adipose-derived mesenchymal stem cells and human lung squamous cell carcinoma cells, respectively.

We present an efficient algorithm to compute multidimensional spatially variant convolutions--or inner products--between N-dimensional signals and B-splines--or their derivatives--of any order and arbitrary sizes. The multidimensional B-splines are computed as tensor products of 1-D B-splines, and the input signal is expressed in a B-spline basis. The convolution is then computed by using an adequate combination of integration and scaled finite differences as to have, for moderate and large scale values, a computational complexity that does not depend on the scaling factor. To show in practice the benefit of using our spatially variant convolution approach, we present an adaptive noise filter that adjusts the kernel size to the local image characteristics and a high sensitivity local ridge detector.

It has been shown that employing multiple atlas images improves segmentation accuracy in atlas-based medical image segmentation. Each atlas image is registered to the target image independently and the calculated transformation is applied to the segmentation of the atlas image to obtain a segmented version of the target image. Several independent candidate segmentations result from the process, which must be somehow combined into a single final segmentation. Majority voting is the generally used rule to fuse the segmentations, but more sophisticated methods have also been proposed. In this paper, we show that the use of global weights to ponderate candidate segmentations has a major limitation. As a means to improve segmentation accuracy, we propose the generalized local weighting voting method. Namely, the fusion weights adapt voxel-by-voxel according to a local estimation of segmentation performance. Using digital phantoms and MR images of the human brain, we demonstrate that the performance of each combination technique depends on the gray level contrast characteristics of the segmented region, and that no fusion method yields better results than the others for all the regions. In particular, we show that local combination strategies outperform global methods in segmenting high-contrast structures, while global techniques are less sensitive to noise when contrast between neighboring structures is low. We conclude that, in order to achieve the highest overall segmentation accuracy, the best combination method for each particular structure must be selected.

In this paper we present an adaptive B-spline smoothing algorithm to reduce noise in biomedical images. The filter consists of a convolution between the input image and an adaptive B-spline kernel that varies its size according to the local image characteristics. We show that this filter can be efficiently implemented by taking advantage of the convolution and factorization properties of B-splines. We tested our filter on phantoms and real images, obtaining very high noise reduction in homogeneous areas with little degradation of the edges.

Statistical analysis of genetic changes within cell nuclei that are far from the primary tumor would help determine whether such changes have occurred prior to tumor invasion. To determine whether the gene amplification in cells is morphologically and/or genetically related to the primary tumor requires quantitative evaluation of a large number of cell nuclei from continuous meaningful structures such as milk-ducts, tumors, etc., located relatively far from the primary tumor. To address this issue, we have designed an integrated image analysis software system for high-throughput segmentation of nuclei. Filters such as Beltrami flow-based reaction-diffusion, directional diffusion, etc., were used to pre-process the images resulting in a better segmentation. The accurate shape of the segmented nucleus was recovered using an iterative "shrink-wrap" operation. The study of two cases of ductal carcinoma in situ in breast tissue supports the biological observation regarding the existence of a preferential intraductal invasion, and therefore a common origin, between the primary tumor and the gene amplification in the cell-nuclei lining the ductal structures in the breast.

Spatial events largely determine the biology of cells, tissues, and organs. In this paper, we present a tool for the quantitative spatial analysis of heterogeneous cell populations, and we show experimental validation of this tool using both artificial and real (mammary gland tissue) data, in two and three dimensions. We present the refined relative neighborhood graph as a means to establish neighborhood between cells in an image while modeling the topology of the tissue. Then, we introduce the M function as a method to quantitatively evaluate the existence of spatial patterns within one cell population or the relationship between the spatial distributions of multiple cell populations. Finally, we show a number of examples that demonstrate the feasibility of our approach.