Using histopathology latent diffusion models as privacy-preserving dataset augmenters improves downstream classification performance.

Latent diffusion models (LDMs) have emerged as a state-of-the-art image generation method, outperforming previous Generative Adversarial Networks (GANs) in terms of training stability and image quality. In computational pathology, generative models are valuable for data sharing and data augmentation. However, the impact of LDM-generated images on histopathology tasks compared to traditional GANs has not been systematically studied. We trained three LDMs and a styleGAN2 model on histology tiles from nine colorectal cancer (CRC) tissue classes. The LDMs include 1) a fine-tuned version of stable diffusion v1.4, 2) a Kullback-Leibler (KL)-autoencoder (KLF8-DM), and 3) a vector quantized (VQ)-autoencoder deploying LDM (VQF8-DM). We assessed image quality through expert ratings, dimensional reduction methods, distribution similarity measures, and their impact on training a multiclass tissue classifier. Additionally, we investigated image memorization in the KLF8-DM and styleGAN2 models. All models provided a high image quality, with the KLF8-DM achieving the best Frechet Inception Distance (FID) and expert rating scores for complex tissue classes. For simpler classes, the VQF8-DM and styleGAN2 models performed better. Image memorization was negligible for both styleGAN2 and KLF8-DM models. Classifiers trained on a mix of KLF8-DM generated and real images achieved a 4% improvement in overall classification accuracy, highlighting the usefulness of these images for dataset augmentation. Our systematic study of generative methods showed that KLF8-DM produces the highest quality images with negligible image memorization. The higher classifier performance in the generatively augmented dataset suggests that this augmentation technique can be employed to enhance histopathology classifiers for various tasks.

A simple approach for detecting HLA-A*02 alleles in archival formalin-fixed paraffin-embedded tissue samples and an application example for studying cancer immunoediting.

The HLA system represents a central component of the antigen presentation machinery. As every patient possesses a defined set of HLA molecules, only certain antigens can be presented on the cell surface. Thus, studying HLA type-dependent antigen presentation can improve the understanding of variation in susceptibility to various diseases, including infectious diseases and cancer. In archival formalin-fixed paraffin-embedded (FFPE) tissue, the HLA type is difficult to analyze because of fragmentation of DNA, hindering the application of commonly used assays that rely on long DNA stretches. Addressing these difficulties, we present a refined approach for characterizing presence or absence of HLA-A*02, the most common HLA-A allele in the Caucasian population, in archival samples. We validated our genotyping strategy in a cohort of 90 samples with HLA status obtained by an NGS-based method. 90% (n = 81) of the samples could be analyzed with the approach. For all of them, the presence or absence of HLA-A*02 alleles was correctly determined with the method, demonstrating 100% sensitivity and specificity (95% CI: 91.40%-100% and 91.19%-100%). Furthermore, we provide an example of application in an independent cohort of 73 FFPE microsatellite-unstable (MSI) colorectal cancer samples. As MSI cancer cells encompass a high number of mutations in coding microsatellites, leading to the generation of highly immunogenic frameshift peptide antigens, they are ideally suited for studying relations between the mutational landscape of tumor cells and interindividual differences in the immune system, including the HLA genotype. Overall, our method can help to promote studying HLA type-dependency during the pathogenesis of a wide range of diseases, making archival and historic tissue samples accessible for identifying HLA-A*02 alleles.

Deep learning detects genetic alterations in cancer histology generated by adversarial networks.

Deep learning can detect microsatellite instability (MSI) from routine histology images in colorectal cancer (CRC). However, ethical and legal barriers impede sharing of images and genetic data, hampering development of new algorithms for detection of MSI and other biomarkers. We hypothesized that histology images synthesized by conditional generative adversarial networks (CGANs) retain information about genetic alterations. To test this, we developed a 'histology CGAN' which was trained on 256 patients (training cohort 1) and 1457 patients (training cohort 2). The CGAN synthesized 10 000 synthetic MSI and non-MSI images which contained a range of tissue types and were deemed realistic by trained observers in a blinded study. Subsequently, we trained a deep learning detector of MSI on real or synthetic images and evaluated the performance of MSI detection in a held-out set of 142 patients. When trained on real images from training cohort 1, this system achieved an area under the receiver operating curve (AUROC) of 0.742 [0.681, 0.854]. Training on the larger cohort 2 only marginally improved the AUROC to 0.757 [0.707, 0.869]. Training on purely synthetic data resulted in an AUROC of 0.743 [0.658, 0.801]. Training on both real and synthetic data further increased AUROC to 0.777 [0.715, 0.821]. We conclude that synthetic histology images retain information reflecting underlying genetic alterations in colorectal cancer. Using synthetic instead of real images to train deep learning systems yields non-inferior classifiers. This approach can be used to create large shareable data sets or to augment small data sets with rare molecular features. © 2021 The Authors. The Journal of Pathology published by John Wiley & Sons, Ltd. on behalf of The Pathological Society of Great Britain and Ireland.

The shared frameshift mutation landscape of microsatellite-unstable cancers suggests immunoediting during tumor evolution.

The immune system can recognize and attack cancer cells, especially those with a high load of mutation-induced neoantigens. Such neoantigens are abundant in DNA mismatch repair (MMR)-deficient, microsatellite-unstable (MSI) cancers. MMR deficiency leads to insertion/deletion (indel) mutations at coding microsatellites (cMS) and to neoantigen-inducing translational frameshifts. Here, we develop a tool to quantify frameshift mutations in MSI colorectal and endometrial cancer. Our results show that frameshift mutation frequency is negatively correlated to the predicted immunogenicity of the resulting peptides, suggesting counterselection of cell clones with highly immunogenic frameshift peptides. This correlation is absent in tumors with Beta-2-microglobulin mutations, and HLA-A*02:01 status is related to cMS mutation patterns. Importantly, certain outlier mutations are common in MSI cancers despite being related to frameshift peptides with functionally confirmed immunogenicity, suggesting a possible driver role during MSI tumor evolution. Neoantigens resulting from shared mutations represent promising vaccine candidates for prevention of MSI cancers.

Spatially resolved electrical impedance methods for cell and particle characterization.

Electrical impedance is an established technique used for cell and particle characterization. The temporal and spectral resolution of electrical impedance have been used to resolve basic cell characteristics like size and type, as well as to determine cell viability and activity. Such electrical impedance measurements are typically performed across the entire sample volume and can only provide an overall indication concerning the properties and state of that sample. For the study of heterogeneous structures such as cell layers, biological tissue, or polydisperse particle mixtures, an overall measured impedance value can only provide limited information and can lead to data misinterpretation. For the investigation of localized sample properties in complex heterogeneous structures/mixtures, the addition of spatial resolution to impedance measurements is necessary. Several spatially resolved impedance measurement techniques have been developed and applied to cell and particle research, including electrical impedance tomography, scanning electrochemical microscopy, and microelectrode arrays. This review provides an overview of spatially resolved impedance measurement methods and assesses their applicability for cell and particle characterization.

Self-Learning Microfluidic Platform for Single-Cell Imaging and Classification in Flow.

Single-cell analysis commonly requires the confinement of cell suspensions in an analysis chamber or the precise positioning of single cells in small channels. Hydrodynamic flow focusing has been broadly utilized to achieve stream confinement in microchannels for such applications. As imaging flow cytometry gains popularity, the need for imaging-compatible microfluidic devices that allow for precise confinement of single cells in small volumes becomes increasingly important. At the same time, high-throughput single-cell imaging of cell populations produces vast amounts of complex data, which gives rise to the need for versatile algorithms for image analysis. In this work, we present a microfluidics-based platform for single-cell imaging in-flow and subsequent image analysis using variational autoencoders for unsupervised characterization of cellular mixtures. We use simple and robust Y-shaped microfluidic devices and demonstrate precise 3D particle confinement towards the microscope slide for high-resolution imaging. To demonstrate applicability, we use these devices to confine heterogeneous mixtures of yeast species, brightfield-image them in-flow and demonstrate fully unsupervised, as well as few-shot classification of single-cell images with 88% accuracy.