A study on automated image analysis for cancer diagnostics
Cancer diagnosis has remained largely unchanged for decades and requires a pathologist (a specialist doctor) to look at stained tissue slices removed following surgery or biopsy under a microscope. By looking at the appearance of the material (so-called ‘morphology’), the pathologist can identify a particular cancer, where present. In turn, this determines the treatment open to the patient. Very recently, pathology has benefited from advances in computing: tissue slices are scanned to generate high-resolution images of tissue (much the same as one might do with a paper document). Pathologists can then examine these images on a computer rather than on a microscope.
An additional benefit of this new technology is that these images could be used to develop computer–based analyses which can look at the tissue morphology in a manner similar to what the pathologist would be doing when making a diagnosis. This strategy does not aim to replace pathologists, but it has the potential to increase the speed at which they make a diagnosis, as well as its accuracy. In turn, this would shorten the time taken for a pathologist to review and report a case, meaning that patients and their clinical care team could receive the diagnosis sooner. In partnership with funders (4Dpath), we have developed a computer-based processing system that can identify a range of cancers without human input. This system extracts information from the image that it uses to achieve a diagnosis. This system can differentiate between malignant (cancerous) and non-malignant (benign) specimens. However, it can also - unlike other existing systems - provide information on cancer type, grade (how aggressive the cancer appears), and whether it has spread to the bloodstream.
The first aim of this project is to collect a large database of images with which to refine and test this computer-based system so that it can diagnose a wide range of cancers (e.g. breast, skin, digestive system, liver, reproductive system, lung, brain and urinary system). This will involve scanning 12,000 cases, mostly cancers but also other non-cancer diagnoses for comparison.
The second aim of this project is to match image data with patient’s own personal features, such as age, general health, treatment, disease recurrence (the cancer ‘coming back’) and survival using complex mathematics. We hope that this may help to give patients a more accurate outlook on how they are likely to respond to treatment. This involves accessing patient records to collect information about disease and response to treatment, although once collected, all data are anonymised (i.e. all identifiable patient information is removed). Patients will however not be asked for permission to access their records since this project is based on old, historical cases, where many patients will unfortunately have passed away. However, we recognise that some patients will still be alive and may be unwilling to have their data used for research. Therefore, we invite anyone who would like to know more about this project or who would wish to ensure that their records are not accessed for research to contact us by 30th October 2018. This will enable us to keep a register of names to prevent such data to be used.
The project has been approved by a Patient & Public Involvement Group, a Research Ethics Committee and the Confidentiality Advisory Group and is fully compliant with the legal framework.
For enquiries and further details, please contact the Chief Investigator:
Dr Nic Orsi
Leeds Institute of Cancer & Pathology
Wellcome Trust Brenner Building
St James’s University Hospital
Beckett Street
Leeds LS9 7TF
Tel: 0113 3438625
Email: mednmo@leeds.ac.uk
