Tech giant Google has once again demonstrated the breadth of its healthcare capabilities, this time successfully creating a deep learning AI algorithm capable of diagnosing breast cancer.

In an experiment by Google Research scientists as part of its Brain Residency Program (an educational course in machine and deep learning), an AI algorithm was trained to analyse digital pathology slides and pick out cancerous tumours.

Once the team had ‘trained’ the algorithm, the algorithm achieved 92% sensitivity in a minimal timeframe – significantly higher than the 73% achieved by trained pathologists with no time constraint.

Additionally, the team was able to recreate the algorithm’s accuracy in other datasets taken from different hospitals and scanning machines.

The area of diagnosis through image analysis is one many are expecting to be disrupted by AI. After all, Google has been relying on a deep learning algorithm for years to provide accurate image search results in its self-titled search engine.

In general practice, pathology analysis can be a long and time-consuming process particularly in regards to the average patient having 10 or more pathology slides for their tumour.

Even once the analysis process is complete, the diagnosis is not always definitive and the degree of disagreement between trained pathologists is high.

Although Google Research is evidently investigating deep learning algorithms in healthcare settings, its most notable effort in the field is its DeepMind Health branch. The division is currently working with several NHS trusts to see how AI can improve care, including with Moorfields Eye hospital and the University College London to improve the diagnose of diabetic retinopathy and age-related macular degeneration, and to help plan radiotherapy, respectively.

Other AI specialists like Israeli startup Zebra Medical have already developed a catalogue of algorithms capable of diagnosing numerous conditions, including several types of cancer and bone fractures.

A deep learning algorithm was recently shown to be as effective as dermatologists at distinguishing between benign and malignant skin growths.