Steve meade designs tm-1c
Ongoing research is investigating associations between clinical interventions and CMS subgrouping, and so this classification has the potential to guide treatment allocation in future clinical practice. 12 An association of specific imaging features with meta-gene expression profiles was described and preliminary data were subsequently reported indicating that histology slides may contain sufficient information to predict the CMS molecular subtypes of CRC. Prior studies have shown the feasibility of developing image-based biomarkers for molecular subtypes of CRC by deep learning. 1 11 Samples with transitioning phenotypes or intratumoural heterogeneity are presently considered to be unclassifiable (13%). These include CMS1 (14% microsatellite instability immune, favourable prognosis in early-stage disease, adverse prognosis in the metastatic setting), CMS2 (37% canonical, epithelial gene expression profile, WNT and MYC signalling activation, intermediate prognosis), CMS3 (13% epithelial profile with evident metabolic dysregulation, intermediate prognosis) and CMS4 (23% mesenchymal, prominent transforming growth factor-β activation, poor prognosis). 11 The CMS classification distinguishes four groups of CRC with distinct clinical behaviour and underpinning biology. 10 Tumour and non-tumour tissue contributes to image information on the histological slide and to the consensus molecular subtype (CMS) classification of CRC at the transcriptional level. 9 The composition of the tumour microenvironment is a key component determining the tumour progression and therapy response. In CRC, it is well known that tumour morphology, growth pattern and architecture hold important clues to differentiating biological subtypes with clinical impact.
Given that H&E processing allows analysis of large tissue sections within existing clinical workflows, the discovery of morpho-molecular correlations holds the promise of improving patient stratification in clinical practice through the development of new image-based biomarkers. Complex multi-scale morphological traits as well as genomic alterations can now be characterised at scale and with short turnaround times. By combining an image-based analysis with molecular characterisation, it is now feasible to identify novel genotype–phenotype correlations.
Steve meade designs tm 1c driver#
Coudray et al 7 use this approach to detect targetable oncogenic driver mutations in lung cancer using deep neural classification networks. 6 More recently, deep learning is being used to capture morphological differences with a precision that exceeds human performance. The application of traditional image analysis to histopathology facilitates the quantitative assessment of tissue architecture, cell distribution and cellular morphology by light microscopy to generate feature libraries of unprecedented resolution and detail. 4 5 In contrast, histopathology slides are inexpensive to produce and principal stains such as H&E are firmly established in the pathology laboratory. 3 Next generation sequencing technologies enable the multi-omic profiling of malignant tumours but mutation and copy number data have been of limited impact in CRC, while more informative RNA analyses are more costly, difficult to standardise and require data storage and bioinformatics expertise. 2 Molecular stratification of patients with CRC is essential to form homogeneous subgroups for targeted treatment and prognosis. 1 An increasing understanding of CRC biology has led to the development of targeted treatments directed against key pro-oncogenic signalling pathways, but these treatments are only effective in a small proportion of patients. Colorectal cancer (CRC) is a disease with heterogeneous molecular subtypes, variable clinical course and prognosis.