Non-disruptive collagen characterization in clinical histopathology using cross-modality image synthesis

Abstract

The importance of fibrillar collagen topology and organization in disease progression and prognostication in different types of cancer has been characterized extensively in many research studies. These explorations have either used specialized imaging approaches, such as specific stains (e.g., picrosirius red), or advanced and costly imaging modalities (e.g., second harmonic generation imaging (SHG)) that are not currently in the clinical workflow. To facilitate the analysis of stromal biomarkers in clinical workflows, it would be ideal to have technical approaches that can characterize fibrillar collagen on standard H&E stained slides produced during routine diagnostic work. Here, we present a machine learning-based stromal collagen image synthesis algorithm that can be incorporated into existing H&E-based histopathology workflow. Specifically, this solution applies a convolutional neural network (CNN) directly onto clinically standard H&E bright field images to extract information about collagen fiber arrangement and alignment, without requiring additional specialized imaging stains, systems or equipment.

Mueller matrix imaging for collagen scoring in mice model of pregnancy

“…The circumferential arrangement of the collagen fibers is also confirmed with the SHG microscopy measurements 24 (Fig. 1 d) that is the gold standard technique for collagen visualization 25 , 26 . SHG microscopy has been used extensively to image the mouse cervix which is reach of collagen type 1.…”

Real-time polarization microscopy of fibrillar collagen in histopathology

“…However, due to need of fluorescence imaging and the computational approach employed, this method can’t be integrated into every pathologists workflow without disrupting current practice. To remove the need of the optical system we have recently developed a convolutional neural network-based method that can generate virtual SHG images directly from bright-field images of H&E-stained slides after training with breast and pancreatic tissue sections 62 , 63 , but such models must be verified before use on tissue types outside the training pool. However, PPM can visualize and characterize collagen fibers in H&E-stained slides, aided by the optically generated color signals, and thus provides a unique opportunity for incorporating stromal biomarkers in diagnosis and prognosis.…”

Machine-Learning–Based Evaluation of Intratumoral Heterogeneity and Tumor-Stroma Interface for Clinical Guidance

“…40 However, despite the accumulated evidence of the informative value of both explicit and implicit collagen microarchitecture, clinical adoption of the models is delayed by the lack of affordable high-capacity workflows of histology processing, imaging, and computation. Recently, computational methods for extracting collagen characteristics from routine H&E sections were proposed by Keikhosravi et al, 41 who trained a convolutional neural network model on SHG data to synthesize SHG images from H&E images. Results were consistent with the SHG ground truth, and the signal generation method can potentially compensate for some of the limitations (orientation dependence and sensitivity) of SHG imaging itself.…”