Application of Brisbane developed technology has created ‘data stability’ at population scale in a world-first study into the incidence of cancer in 45,000 patients. The landmark approach using machine learning for clinical insight is reported in prestigious global Oncology journal
In what is described as a world first, Max Kelsen, an AI and machine learning consultancy, has applied machine learning to help researchers at the QIMR Berghofer Medical Research Institute gain unprecedented insights into the incidence and diagnoses of skin cancers in Australia.
Since 2011, QIMR Berghofer researchers have been engaged in a 20-year longitudinal study to discover the incidence of non-melanoma skin cancers in Australia. The ‘QSkin’ study is the largest research study ever conducted on skin cancer and is considered a landmark project to inform treatment, and government policy, on skin cancer. Skin cancer affects more Australians than any other nation (per capita).
According to David Whiteman, deputy director of QIMR Berghofer “The QSkin study is the largest study of skin cancer in the world. We keep track of skin cancer events in QSKIN by working closely with pathology
companies around Queensland. The pathologists send reports to us relating to skin cancers among
He said, “For the first few years of the study, QSKIN investigators read and processed every single report (almost 30 000) to extract relevant details about the skin cancer diagnoses, and then manually entered all of the information into a database. The web application developed by Max
Kelsen is trained to automatically retrieve data about each diagnosis, and can do this in a fraction of the time that it would take our staff to achieve the same result.”
The project is focused on discovering the incidence of improving the diagnosis and classification of Keratinocyte cancers (previously non-melanoma skin cancers). These include basal and squamous cell carcinomas, which are exceedingly common in high-risk populations, but for which accurate measures of their incidence are seldom derived, due partly to the burden of manually reviewing pathology reports required to extract relevant diagnostic information.
The QIMR Berghofer study, its researchers were faced with the additional challenge of extracting relevant data from an enormously large and complex data set from 45,000 Queenslanders, from the ages of 40-69, all of whom are volunteers.
Brisbane based Max Kelsen applies artificial intelligence (AI) and machine learning (ML) to improve competitiveness in commercial organisations and accelerate vital discoveries in human life sciences.
It began working with QIMR Berghofer in 2018 to help QIMR researchers uncover faster and more accurate ways to map the incidence of rare skin cancers by developing a Web-based, machine learning application capable of processing large numbers of pathology reports.
By applying a set of machine learning-based rules over the massive data set, an unprecedented level of ‘data stability’ was achieved using ML, leading to greater accuracy of cancer type classifications amongst individuals across age, place of living, sun exposure, and even genetic predisposition. The company claims this is the first time a model of this type has been established and deployed in a working clinical setting in Australia, saving QIMR Berghofer thousands of hours of human analysis and freeing up clinicians from the monotonous task of data entry, to focus on their core skill set.
By leveraging transformative technology, QIMR Berghofer can now be confident of a scalable application capable of processing the increasing amounts of data up until the end of the study in 2031 without fear of increased operating costs and human error.
Cameron Bean, Global Health and Life Science Practice Lead, Max Kelsen, said “QIMR Berghofer is leading the world in its research, and it is exciting to be able to apply machine learning in such a prestigious and patient-focused study, based in Brisbane but with global impact.”
The study achieved global recognition earlier this year when it was published in the prestigious, US-based JCO Clinical Cancer Informatics journal. The report concluded “Supervised learning methods were used to develop a Web application capable of accurately and rapidly classifying large numbers of pathology reports for keratinocyte cancers and related diagnoses. Such tools may provide the means to accurately measure subtype-specific skin cancer incidence.”