Leaders from Europe and the US convened to explore exciting leaps forward of AI in oncology at the HIMSS21 & Health 2.0 European Health Conference, though the panel highlighted key barriers to greater acceptance and adoption of AI into mainstream care.
The ‘New Frontiers of AI and Data Analytics in Oncology’ session, moderated by Professor Karol Sikora, chief medical officer, Rutherford Health and former chief of the Cancer Programme, WHO, also saw leaders share innovative applications for AI used across the cancer pathway. The panel of experts also included Professor Barbara Alicja Jereczek-Fossa, associate professor of Radiation Oncology, University of Milan and head of Radiotherapy Division, European Institute of Oncology, and her colleague, Eng. Matteo Pepa, biomedical engineer, Division of Radiation Oncology, IEO. Joining from the US was Dr Tufia Haddad, chair of Digital Health, Department of Oncology, Mayo Clinic and chair of Practice Innovation and Platform, Mayo Clinic Cancer Center.
WHY IT MATTERS
While AI is already widely used in oncology in image analysis and other areas, exciting new applications are being trialled at leading cancer centres across the globe. This includes research into ever-more advanced AI-tools to support clinical workflow and to identify biomarkers for early disease detection.
Barriers to acceptance and wider adoption of AI included a need to make AI-powered decisions more ‘explainable’ to clinicians, getting over the ‘black box’ problem, and to improve the quality of data for algorithm training.
While Sikora pointed out cancer has in many ways led the way for AI in healthcare, COVID-19 has prompted its wider deployment, with the use of AI algorithms to help clear the patient backlog and to pinpoint COVID-19 hotspots. Going forward, AI will be increasingly relied on to draw actionable insights from expanding volumes of patient data, integrated into the EHR from multiple sources. Ultimately, AI – with its ability to detect patterns faster than the human eye – will be central to the drive for greater personalisation of healthcare and an acceleration in new treatments and cures. However Goyen reinforced that AI must “complement and not replace clinicians.”
ON THE RECORD
The session took a deep dive into cutting-edge applications for AI in oncology at all stages of the patient journey, as well as looking to future use.
Goyen shared examples of AI use in diagnostic radiology to streamline workflow and support radiologists to create richer, more definitive reports translating into “more informed clinical decision with higher diagnostic confidence”.
He covered a workflow management system that simplifies the selection, deployment and usage of AI in the PACS imaging workflows, and a tool that can cut the reading time of a 3D tomosynthesis scan for breast cancer by an average of 30%.
Goyen commented: “Ideally, radiologists don’t even know that AI is working, they just realise that their work list is prioritised, and that their tasks are completed more efficiently.”
Next, Jereczek-Fossa and Eng. Pepa talked about the role of AI, machine learning and deep learning in radiation oncology in reducing planning time and improving image quality.
Finally, Haddad shared insights about a clinical trial-matching tool her clinic is developing to save clinicians time in mining multiple sources of data. She said: “We leveraged NLP and machine learning to really train the system. First to identify patient and tumour attributes that are needed in order to match to the inclusion and exclusion criteria present in an individual clinical trial amongst a portfolio of clinical trials.”
Introducing the tool into clinical practice led to a “dramatic increase” in clinical trial accruals. “There was an immediate and sustained impact in terms of how this system allowed our providers to better understand the clinical trial opportunities. And because they were doing it more efficiently, they now had more time to spend with the patient, counselling them about the potential benefits of participating in a clinical trial,” Haddad concluded.