Automated Cancer Diagnosis and Prognosis Using Digital Images (v. 1)
Research type
Research Study
Full title
Automation of Cancer Diagnosis and Prognosis Using Analysis Digital Images of Scanned Slides
IRAS ID
231039
Contact name
Nicolas M Orsi
Contact email
Duration of Study in the UK
9 years, 11 months, 31 days
Research summary
Cancer diagnosis has remained largely unchanged for decades and requires a pathologist (specialist diagnostician doctor) to look at stained sections of tissue removed at surgery/biopsy under a microscope in order to identify a disease entity. However, recent advances in microscopy have allowed the digital capture of high resolution, high magnification images of entire sections of tissue (i.e. scanning slides, much the same as one might do with a paper document). This has opened up the possibility of developing computer analytical methods which, when applied to these images, could increase diagnostic speed and improve specimen turnaround times for patient benefit (thus overcoming the diagnostic time taken by a pathologist to review and report a case). In partnership with funders (4Dpath), we have developed a machine learning algorithm (a computer-based processing system) which can identify a range of cancers of both skin and breast without human input. This algorithm extracts morphological features from images (i.e. their appearance) which it uses to achieve a diagnosis. This system cannot only differentiate between malignant (cancerous) and non-malignant (benign) specimens, it can also - unlike other existing labour intensive approaches - provide information on cancer subtype, grade (how aggressive the cancer appears), vascular invasion (whether it has spread to the bloodstream) and (to some degree) prognosis.
The aim of this project is to collect a large database of images with which to refine and validate this algorithm so that it can diagnose a wider range of cancers of the breast, skin, digestive system, liver, female reproductive tract, lung, urinary tract and hormonal system. This will involve scanning 12,000 cases of both malignant and benign diagnoses. The data collected will be fused with clinico-demographic patient information (i.e. age, treatment received, survival) using complex mathematics to further optimise and validate its predictive (prognostic) capability on patient clinical outcome.REC name
Wales REC 5
REC reference
18/WA/0222
Date of REC Opinion
30 Aug 2018
REC opinion
Further Information Favourable Opinion