What Google plans to do with the 700 cancer scans it just acquired

Google has acquired 700 anonymized CT and MRI scans of cancer patients for the purpose of analyzing the documents using technologies developed in its artificial intelligence division, DeepMind.

The company attained the scans via a partnership with the University College London Hospitals (UCHL) NHS Foundation Trust. Both sides hope that the technology will help speed up planning for oral cancer treatments, as the process is slow due to the extreme care and precision with which radiotherapy treatments need to be planned out. 

This is due to the fact that radiotherapy, specifically, zaps cancerous cells rather than the surrounding cells. As such, it’s imperative that the delivery of the treatment is done without damaging nearby nerves and organs.

UCHL will work with researchers from DeepMind to see if machine learning can cut down the planning time for these treatments; specifically, the “segmentation” process. This is when the clinicians create a detailed map of all the areas that need to be treated. 

On the DeepMind side, the computers will be fed data from these scans to draw their own inferences, without human aid. The hope is to develop an algorithm that the machines can use to quickly differentiate cancerous cells from those deemed healthy. 

“Clinicians will remain responsible for deciding radiotherapy treatment plans but it is hoped that the segmentation process could be reduced from up to four hours to around an hour,” DeepMind said in a statement.

For those curious or otherwise concerned about what will happen to the information gleaned from this project, it’s worth pointing out that all data produced throughout this project will be retained by the hospital. Additionally, Google has agreed that it will have its DeepMind division “securely destroy” any information it has on its side. 

All patients whose medical information will be used in this project have previously consented to allowing their scans be used for medical research. 

To learn more, visit DeepMind