A Novel Deep Machine Learning Approach to Detecting Tumour Infiltrating Lymphocytes in Testicular Germ Cell Tumours to Create a Reproducible Tool for Personalised Medicine
Clare Verrill, Nina Linder, Johan Lundin, Andrew Protheroe and others
In a collaborative project with the Finnish Institute of Molecular Medicine, Helsinki, a machine learning algorithm was developed, trained and validated to assess tumour infiltrating lymphocytes (TILs) on H&E whole slide images in testicular tumours. The algorithm was then run on 100 testicular germ cell tumour cases. TIL count was measured in association with various clinical parameters such as stage at presentation and relapse. This was compared with the predictive value of pathologist assessment of TILs in association with the same parameters.
This project builds on work being undertaken by Verrill and Protheroe (co-lead and lead of 100,000 Genomes Project Genomic Clinical Interpretation Partnership (GeCIP))
There is also a second collaborative project with Dr Clare Turnbull at ICR, where the algorithm will be utilised to characterise TILs in a cohort of chemotherapy resistant testicular tumours which have undergone whole exome sequencing.