Machine Learning for Histology
of tumors from children of all nations
What if we could ensure the correct diagnosis of children in every country of the world, whether in a developing country or a rural area of a developed country? Childhood cancers are rare, and the expertise is even more rare. Technology can help.
Our goals are:
Create an international registry where patients upload a simple, digitized version of a child’s cancer’s microscope slide.
Expert pathologists will review every slide, to ensure the correct diagnosis.
The expert will train the machine learning engine to recognize each version of childhood cancer, within 97% accuracy of diagnosis.
Academic and nonprofit hospitals will also provide DNA and RNA sequencing data to accompany their histology slide scans. These special (multidimensional) datasets will be used to understand what mutations are responsible for special sub-features of the histology for each kind of cancer (for example, RB1 gene mutation leads to “anaplasia”).
1. When tightly linked, the presence of either the gene change or histology feature would be used to inform better treatment for a given patient, using a very specific medicine regimen.
2. For developing countries, where only histology slide scans exist, the machine learning inferences of probable mutations in a child’s cancer could be used to infer what medicines might be best for that patient; specifically, for relapsed patients this introduces the opportunity for personalized medicine with compassionate use drug access from merely a histology slide.
Reference: Classification and mutation prediction from non–small cell lung cancer histopathology images using deep learning, Nature Medicine (2018)
Machine Learning — What's My Diagnosis?
Fill out the form below, and on the next page, upload a de-identified soft tissue sarcoma slidescan and let us guess the diagnosis!
Your sample will then be added to the training set.
After submitting this form, you will be redirected to a page where you can upload your de-identified soft tissue sarcoma slidescan.
If you do not upload a file, we will not be able to review and provide a histology-based research, non-clinical diagnosis