Machine learning offers means to raise accuracy of cancer analysis
AI collaborators in Britain and the UK have developed a computer system that they say can spot cancer signs in samples with outstanding levels of accuracy.
Titled ‘Histomorphological Phenotype Learning’ (HPL), the system created by teams from Glasgow and New York universities, has the potential to enable faster, more accurate diagnoses and reliable predictions of patient outcomes.
Co-senior author of the study published in Nature Communications, professor John Le Quesne, from the University of Glasgow’s School of Cancer Sciences, supervised the research.
He admitted: “We were surprised but very pleased by the effectiveness of machine learning to tackle this task.”
Current procedure involves pathologists examining tissue samples taken from cancer patients on microscope slides under a microscope with their observations helping determine patient treatment and assessing chances of recovery.
“It takes many years to train human pathologists to identify the cancer subtypes they examine under the microscope and draw conclusions about the most likely outcomes for patients,” he explained.
“It’s a difficult, time-consuming job, and even highly-trained experts can sometimes draw different conclusions from the same slide.”
The study began with thousands of high-resolution images of tissue samples of lung adenocarcinoma taken from 452 patients, held in the United States National Cancer Institute’s Cancer Genome Atlas database, some of which contains information on how patient cancers progressed.
An algorithm was developed that employed the training process self-supervised deep learning to analyse the images and spot patterns based solely on visual data.
It reduced each slide image into thousands of tiny tiles that were scrutinised by a deep neural network able to teach itself to recognise and classify visual features shared across any of the cells in each sample.
Senior author and research supervisor Dr Ke Yuan of Glasgow’s School of Computing Science revealed: “We didn’t provide the algorithm with any insight into what the samples were or what we expected it to find. Nonetheless, it learned to spot recurring visual elements in the tiles which correspond to textures, cell properties and tissue architectures called phenotypes.
“By comparing those visual elements across the whole series of images it examined, it recognised phenotypes which often appeared together, independently picking out the architectural patterns that human pathologists had already identified in the samples.”
Analysing slides from squamous cell lung cancer, the HPL system distinguished between their features with 99% accuracy.
Researchers used the algorithm to analyse links between the phenotypes it had classified and the clinical outcomes stored in the database.
This revealed that phenotypes such as tumour cells which are less invasive, or lots of inflammatory cells attacking the tumour, were more common in patients who lived longer after treatment. Others, such as aggressive tumour cells forming solid masses, or regions where the immune system was excluded, were more closely associated with the recurrence of tumours.
Predictions for the likelihood and timing of cancer’s return made by the HPL system were correct 72% of the time – substantially (12.5%) more than human pathologists’ 64% accuracy.
Similar results were obtained for 10 other types of cancers, claimed the study.
Glasgow's School of Cancer Sciences and School of Computing Science research associate, Dr Adalberto Claudio Quiros, also a co-first author of the paper said the work demonstrated the potential of cutting-edge machine learning.
“This kind of self-learning algorithm will only become more accurate as additional data is added, helping it become more fluent in the language of cancer. Unlike humans, it brings no preconceived ideas to its work, so it may even find patterns across the datasets that haven’t been fully explored before,” he commented.
“Ultimately, our aim is to provide doctors and patients with a tool that can help provide them with an improved understanding of their prognosis and treatment.”
Dr Aristotelis Tsirigos and Dr Nicolas Coudray, of New York University’s Grossman School of Medicine and Perlmutter Cancer Centre, were co-senior investigator and co-first author on the paper, respectively. Researchers from New York University, University College London and the Karolinska Institute also contributed to the paper.
The paper is titled ‘Mapping the landscape of histomorphological cancer phenotypes using self-supervised learning on unlabeled, unannotated pathology slides’. Research funding came from the Engineering and Physical Sciences Research Council (EPSRC), the Biotechnology and Biological Sciences Research Council (BBSRC), and the National Institutes of Health.
Pic: Dr Ke Yuan, ProfJohn Le Quesne and Dr Adalberto Claudio Quiros | cancer sample machine learning algorithm (Credit UniGlasgow / Chris James)
Related Content
Gatsby: a benchmark for STEM success
The Gatsby Foundation’s benchmarks are helping transform school careers guidance. Good news for STEM employers, says Rachel Bibby, who explains …
When plasma leaves the laboratory
The orbital laboratory is evolving into something more ambitious, a controlled industrial environment for producing next-generation materials, explains Ed Smith and Andrew Griffiths. …
London Lab Live: A capital investment
Launched last year as part of the Future Labs’ stable, the London show looks here to stay. One year on, …
