Three-dimensional tumour microenvironment reconstruction and tumour-immune interactions' analysis

Abstract

Tumours arise within complex 3D microenvironments, but the routine 2D analysis of tumours often underestimates the spatial heterogeneity. In this paper, we present a methodology to reconstruct and analyse 3D tumour models from routine clinical samples allowing 3D interactions to be analysed at cellular resolution. Our workflow involves cutting thin serial sections of tumours followed by labelling of cells using markers of interest. Serial sections are then scanned, and digital multiplexed data are created for computational reconstruction. Following spectral unmixing, a registration method of the consecutive images based on a prealignment, a parametric and a non-parametric image registration step is applied. For the segmentation of the cells, an ellipsoidal model is proposed and for the 3D reconstruction, a cubic interpolation method is used. The proposed 3D models allow us to identify specific interaction patterns that emerge as tumours develop, adapt and evolve within their host microenvironment. We applied our technique to map tumourimmune interactions of colorectal cancer and preliminary results suggest that 3D models better represent the tumourimmune cells interaction revealing mechanisms within the tumour microenvironment and its heterogeneity.

A digital pathology workflow for the segmentation and classification of gastric glands: Study of gastric atrophy and intestinal metaplasia cases

“…Given the various challenges, several techniques and methods have been developed, based on either hand-crafted or deep learning features. Hand-crafted developed approaches are based on grayscale density, color, texture and shape information extracting low-level or mid-level set of features [8][9][10]. On the other hand, more sophisticated methods [11,12] and deep-learning techniques including convolutional neural networks (CNNs) [13] and visual transformers (VTs) [14,15] have been developed aiming to address medical image challenges by extracting high-level features directly from the data.…”

Multi-scale Deformable Transformer for the Classification of Gastric Glands: The IMGL Dataset

“…Thus, various techniques and methods, based on either hand-crafted or deep learning features, have been developed for histopathological image classification tasks. Hand-crafted developed classification approaches for digital pathology tasks are based on grayscale density, color, texture and shape information [9], [10], [11]. After the extraction of low-level or mid-level set of features, post-processing methods such as dimensionality reduction and a classifier are usually used aiming to assign a classification label to each image [12].…”

Gland segmentation in gastric histology images: detection of intestinal metaplasia

“…The digital medical image segmentation and classification field receives growing attention and has become more and more popular [10]. Thus, various techniques and methods, based on either hand-crafted or deep learning features, have been developed for histopathological image segmentation and classification tasks.…”