DeepMind – the Google subsidiary dedicated to artificial intelligence –and the Moorfields Eye Hospital in London have developed a form of IA which can detect anomalies on retinal scans.
In July 2016, Google announced that it had formed a partnership with the UK National Health Service (NHS) and Moorfields Eye Hospital in London to develop a form of artificial intelligence capable of diagnosing eye diseases. Less than two years on, DeepMind – the Google subsidiary dedicated to IA – has announced that it has achieved that goal.
The researchers working on the project have submitted their results to a medical journal for publication by next year. If their work is validated by the scientific community, DeepMind is looking to begin clinical tests within the next few years, in conjunction with Moorfields and other hospitals, according to reports in French website Numérama.
Three eye diseases that can be identified thanks to deep learning
The artificial intelligence in question was fed with thousands of anonymous 3D retinal scans from Moorfields Eye Hospital, categorised based on symptoms. After assimilating the medical information using deep learning, the DeepMind algorithm is now capable of recognising scans which show glaucoma, diabetic retinopathy and age-related macular degeneration.
"What we have learned with this project will benefit people over the world, and will help bring an end to avoidable blindness," said a delighted Peng Tee Khaw, head of R&D at Moorfields. Dominic King, clinical head at DeepMind Health, was equally satisfied with the progress made, telling the Financial Times: "In certain specific fields such as medical imaging, we are set to make incredible progress in the next few years with artificial intelligence. Machine learning could play a crucial role, detecting with a greater level of sensitivity and accuracy than is currently possible."
Will IA soon be able to detect cancer?
According to Dr King, the DeepMind algorithm has a "generalised" form of artificial intelligence which is capable of analysing other types of image. It could therefore evaluate the radiotherapy scans provided by University College London Hospital or the mammograms at the health centre of Imperial College London.
Analysing documents such as these using IA would save valuable time for UK hospitals which are currently overworked. According to a Google consultant quoted by the Financial Times, labelling x-rays to detect cancers located in the head and neck takes anything between five and six hours.
Potential IA applications such as these and the arrival of big technology players in the health sector is proving to be cause for concern, however. The ICO – which is the UK data protection authority – has deemed that DeepMind having access to the medical files of 1.6 million patients via the NHS is illegal. Since this ruling, DeepMind has created a research body dedicated to the ethical and social consequences of IA.
Contact Allianz Partners
Paris, France, Feb 6, 2017
Allianz Worldwide Partners conducted the first worldwide study on connected health and on personal data protection, using the panel of Apple Watch owners. The study was carried out by [...]