For most diseases, the first step – to treatment and hopefully cure – is obtaining a correct diagnosis. For many conditions, this requires taking a sample of tissue and examining it individually under a microscope. The process requires skilled pathologists trained to identify potential signs of disease or abnormalities, but increasingly demand for their time and services is outstripping the available supply. The IMI-funded Bigpicture project is seeking ways to use artificial intelligence to automate and speed up the process, potentially improving consistency and helping keep pace with the increasing demand for high-quality diagnostics.
The personal touch in healthcare is essential; when people visit their healthcare provider – be it their general practitioner or a clinic – they seek reassurance and solutions. Yet in many conditions – such as cancer screening – the first step is an accurate assessment and, if necessary, a diagnosis that allows treatment to begin.
Pathology is a series of time-consuming processes; biological samples are fixed to a slide, stained and examined individually by experts, under a microscope, for potential abnormalities. Pathologists are trained to look for specific indicators of disease in order to make a diagnosis, allowing the doctor in charge of the patient to decide what treatment – if any – is required.
While such an approach may be effective in terms of results, the concern is that there are simply not sufficient pathologists available to deal with the demand. Screening – such as is done for cervical cancer – produces huge number of samples for assessment.
If part of the assessment could be automated – without any detriment to the quality of the results – it could simultaneously dramatically speed up the process, decrease the workload on pathologists, and cut costs. The benefits in mass screening programmes – such as cervical cancer – could be immense. Artificial intelligence – AI – offers a potential solution. The concept is that AI could learn from the collected knowledge of successfully diagnosed events in order to be able to recognise patterns. It is an attractive idea, with many potential advantages; however, the challenges are substantial.
Identifying, addressing and overcoming these challenges is the objective of the Innovative Medicines Initiative (IMI) Bigpicture project. The project is an inclusive platform for academic institutions, small- and medium-sized enterprises, public organisations, pharmaceutical companies and other partners. The current 46 partners have joined forces, and are examining the practical, legal and ethical issues involved in developing a large repository of high-quality annotated pathology data. Ultimately, Bigpicture will create the opportunity for developing AI tools and algorithms, paving the way for AI applications and for identifying solutions.
As AI relies on huge amounts of data, Bigpicture is ultimately aiming to make millions of clinical and non-clinical whole slide images (WSI) available for AI development. However, as each image can be gigabytes in size, storing this amount of data – and making it available – will require a significant ICT infrastructure. Then there are the considerable legal and ethical issues of privacy and data confidentiality to be solved before researchers can use the data. In particular the use of clinical data (samples from human patients) requires strict regulations, and patient privacy must always be respected.
Launched in early 2021, Bigpicture has already made significant progress in solving these challenges. Earlier this year, the project recorded a significant advance as the first dataset – a set of 80 WSI of 8 cases of melanoma – was uploaded to the platform, following approval by the national ethical board for the creation and submission of anonymous datasets.
As it progresses, Bigpicture will be a major fillip to the nascent field of computational pathology, offering researchers access to GDPR-compliant, high-quality data for the development of AI algorithms. These will support the pathologist in his/her daily routine and set the new norm in pathology for the future.
If successful, AI-driven, computational pathology could lead to advances in other areas such as rare diseases, where accurate diagnosis remains a real challenge. It also has the possibility to accelerate and improve patient management, treatment, and ultimately patient health.
Bigpicture is supported by the Innovative Medicines Initiative, a partnership between the European Union and the European pharmaceutical industry.