Deep Learning based detection of tumor tissue compartments improves prognostic immunoprofiling in muscle-invasive bladder cancer

Posted 11th November 2019 by Liv Sewell
Worldwide, bladder cancer (BC) is the 11th most commonly diagnosed cancer. In men, BC is the 7thmost commonly diagnosed cancer worldwide.[1]Although men are more likely to develop BC than women, women present with more advanced disease and have worse survival rates.[2]
Muscle-invasive bladder cancers (MIBC) are cancers that have grown into or through the muscle layers of the bladder wall.
Dr. Katharina Nekolla and Ansh Kapil and their team applied Deep Learning enabled pathology to better understand prognostic factors in MIBC. Here we review the significance of the research and share their original poster.
Prognosis
The prognosis for patients with MIBC is poor. MIBC is highly aggressive with a high incidence of metastasis. Five-year survival rates vary between 30 and 50%.[3]
Even with recent research and development of novel immunotherapies and surgical innovations, prognosis and mortality rates of MIBC patients have remained unchangeable over the last thirty years.[4]
It is thought that TNM staging may not effectively account for the complex and dynamic nature of the disease.[5]
There is therefore a need to significantly improve patient stratification and earlier definitive treatment for high risk MIBC patients.[6]
Improving prognostic immunoprofiling
Currently, treatment decisions cannot be based on molecular markers.[7] Novel Immuno-Oncology agents (anti PD-1/PD-L1) have demonstrated activity, but only in a subset of bladder cancer patients. The prognostic value of PD-L1 expression appears to be limited, highlighting the need for more robust biomarkers.[8]
Nekolla, Kapil et al. employed deep learning based detection of tumour epithelial (TE) and non-epithelial (NE) compartments to study the densities of PD-L1, CD3, CD8, FOXP3, PD-1 and CD68 positive cells in TE and NE compartments and assessed their impact on survival.
Their study suggests that in MIBC the prognostic value of the density of immune cells in the tumour core is highly dependent on the compartment analysed. It also suggests that the density of FOXP3 positive cells appears to be a robust prognostic factor across the different tumour compartments.
They believe further insights might be generated by analysing local cell densities and co-occurrence of different cell populations.
This study shows the potential of deep learning based analysis for improving prognosis and treatment of MIBC. View their poster below.
Deep Learning based detection of tumor tissue compartments improves prognostic immunoprofiling in muscle-invasive bladder cancer
Join us at the 6th Digital Pathology and AI Congress: Europe to explore more of the latest techniques and applications of Deep Learning and image analysis for pathology and laboratory medicine. This event is likely to sell out soon. Explore the programme and reserve your place.
[1]EAU Guidelines. Edn. presented at the EAU Annual Congress Copenhagen 2018. ISBN 978-94-92671-01-1. Accessed @ https://uroweb.org/wp-content/uploads/EAU-MIBC-Guidelines-2018V2.pdf.
[2]Ibid.
[3]Nicolas Brieu et al. “Automated tumour budding quantification by machine learning augments TNM staging in muscle-invasive bladder cancer prognosis.” Scientific Reports vol. 9,1 5174, 26 Mar, 2019, doi:10.1038/s41598-019-41595-2, p. 1. Accessed @ https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6435679/#CR28.
[4]Ibid.
[5]Ibid.
[6]Ibid.
[7]EAU Guidelines, 2018.
[8]Katharina Nekolla, Ansh Kapil, Nicolas Brieu, Thomas Herz, Moritz Widmaier, Alexei Budco, Dasa Medrikova, Ivan Kanchev, Marco Testori, Jessica Chan, Katrin Schneider, Phil Dundee, Paul Anderson, Nathan Lawrentschuk, Lih-Ming Wong, Phuong Phan, Peter Gibbs, Monika Baehner, Ben Tran, Günter Schmidt. Deep Learning based detection of tumor tissue compartments improves prognostic immunoprofiling in muscle-invasive bladder cancer. Poster Presented at the Digital Pathology & AI Congress: USA, New York City, June 13-14, 2019.
Leave a Reply