Google’s Deep Learning AI Project to Revolutionize Cancer Screening

Properly identifying whether the patient has cancer is no easy task, even for experienced pathologists. In fact, the likelihood of agreement between two pathologists observing the same slides can be as low as 48 per cent and this could be a big problem when time is off the essence. Google’s Deep Learning AI project is taking on the task however and has so far shown results which are surprisingly accurate.

Google is sharpening up its image-recognition system to tackle cancer screening better for several years now as part of their efforts which initially was a venture into autonomous car research looking into road layout and traffic recognition. A pathologist looking through multiple patient slides that are 10+ gigapixels large can take a considerable amount of time which is why artificial intelligence taught to recognize the same patterns as a pathologist could prove to be the fastest and cost-effective solution for diagnosis. 

The AI system will still require some training and examination but according to Google, they have demonstrated that it was possible to train a model that either matched or exceeded the performance of a pathologist who spent a lot more time looking at each slides. In Google’s experiments, the AI system faced off against expert pathologists without a time limit set to determine how accurate both are at assessment. Their results showed the experienced pathologists achieving 73 per cent accuracy while the GoogLeNet AI system scored an 89 per cent accuracy.

Google further adds that this AI system is not meant to displace pathologists entirely but to be used as a tool to aid them in proper diagnosis. It will give pathologists a much better overview of where to possibly start screening to get a more accurate hit rate than just the usual methods alone. It also allows for more accurate measurement of tumor sizes and help detect other system irregularities which can otherwise be overlooked.