Could computers diagnose cancer? Artificial intelligence shown to spot early signs of a tumour with 92 per cent accuracy

  • Machine can sift millions of cells to spot just a handful of malignant ones 
  • AI algorithm was trained using slides of samples of patients lymph nodes
  • Human pathologists can diagnose breast cancer with 96 per cent accuracy
  • When the machine and human combined, accuracy went to 99.5 per cent

Computers could soon be helping to diagnose cancer in patients with the help of artificial intelligence that has been trained to spots the early signs of the disease.

An AI machine capable of accurately diagnosing breast cancer 92 per cent of the time has been developed by researchers.

While it is still not quite as good as human specialists – who are correct 96 per cent of the time – it suggests that AI could soon be used to speed up and improve cancer screening.

Scientists have used machine learning to create an artificial intelligence system capable of diagnosing breast cancer from lymph node biopsies with 92 per cent accuracy (cancer cells in a lymph node pictured). When combined with a human pathologist this accuracy increased to 99.5 per cent

Scientists have used machine learning to create an artificial intelligence system capable of diagnosing breast cancer from lymph node biopsies with 92 per cent accuracy (cancer cells in a lymph node pictured). When combined with a human pathologist this accuracy increased to 99.5 per cent

The system was developed by computer scientists at Harvard Medical School gave a machine learning algorithm slides of lymph nodes from breast cancer patients.

HOW THE COMPUTER LEARNED 

The researchers used a machine-learning algorithm that has been used for a range of applications including speech recognition and image recognition.

The researchers gave the program hundreds of slides in which a pathologist has labeled regions of cancer and regions of normal cells.

Over time it was able to build a computational model to classify cancerous cells.

The team then identified the specific training examples for which the computer is prone to making mistakes and re-trained the computer using greater numbers of the more difficult training examples. 

In this way, the computer's performance continued to improve.   

Advertisement

Over time, after looking at hundreds of slides, the computer learned to spot the biological features that pathologists look for using a microscope to indicate cancer.

Dr Andrew Beck, a pathologist at Harvard Medical School and the Cancer Research Institute at Beth Isreael Deaconess Medical Centre, said: 'Identifying the presence or absence of metastatic cancer in a patient's lymph nodes is a routine and critically important task for pathologists.

'Peering into the microscope to sift through millions of normal cells to identify just a few malignant cells can prove extremely laborious using conventional methods.

'We thought this was a task that the computer could be quite good at – and that proved to be the case.'

The team recently demonstrated their AI pathologist at the International Symposium of Biomedical Imaging, where it won a competition against a number of other systems.

In tests, the AI system was given slides of lymph nodes and asked to determine whether they contained cancer. It got them right 92 per cent of the time.

But the researchers said when they combined the machine with a human pathologist, the accuracy increased to 99.5 per cent.

'Combining these two methods yielded a major reduction in errors,' said Dr Beck.

'There have been many reasons to think that digitizing images and using machine learning could help pathologists be faster, more accurate and make more accurate diagnoses for patients.

'This has been a big mission in the field of pathology for more than 30 years.

'But it's been only recently that improved scanning, storage, processing and algorithms have made it possible to pursue this mission effectively.

'Our results show that what the computer is doing is genuinely intelligent and that the combination of human and computer interpretations will result in more precise and more clinically valuable diagnoses to guide treatment decisions.'

The researchers trained their system using hundreds of slides that had the features of healthy and cancerous cells labelled. Over time it became able to identify potential tumours from slides of lymph node biopsies

The researchers trained their system using hundreds of slides that had the features of healthy and cancerous cells labelled. Over time it became able to identify potential tumours from slides of lymph node biopsies

The team, who have published their findings on the arXiv.org open source journal, trained the computer to distinguish between cancerous tumor regions and normal regions using a deep multilayer convolutional network.

Their machine-learning algorithims were based on those that have been used for a range of applications including speech recognition and image recognition.

Dr Dayong Wang, who was part of the research team, said: 'In our approach, we started with hundreds of training slides for which a pathologist has labeled regions of cancer and regions of normal cells.

'We then extracted millions of these small training examples and used deep learning to build a computational model to classify them.'

The team then identified the specific training examples for which the computer is prone to making mistakes and re-trained the computer using greater numbers of the more difficult training examples. In this way, the computer's performance continued to improve.

They are not the only group to be using machine learning to help improve cancer diagnosis.

Researchers believe AI may help to speed up the diagnosis of breast cancer and improve the accuracy (mammogram pictured). It currently takes pathologists hours to sift through the millions of healthy cells under a microscope to find cancerous cells in lymph node biopsies

Researchers believe AI may help to speed up the diagnosis of breast cancer and improve the accuracy (mammogram pictured). It currently takes pathologists hours to sift through the millions of healthy cells under a microscope to find cancerous cells in lymph node biopsies

Professor Paul Rees, an engineer at Swansea University, is using image recognition software to teach a computer to identify features in cells that indicate they may be cancerous.

They are working with doctors to build a system capable of diagnosing leukaemia.

Dr Jeroen van der Laak, who leads a digital pathology research group at Radboud University Medical Center in the Netherlands and was an organizer for the competition, said: 'When we started this challenge, we expected some interesting results.

'The fact that computers had almost comparable performance to humans is way beyond what I had anticipated.

'It is a clear indication that artificial intelligence is going to shape the way we deal with histopathological images in the years to come.'

The comments below have not been moderated.

The views expressed in the contents above are those of our users and do not necessarily reflect the views of MailOnline.

We are no longer accepting comments on this article.