26 April 2023
Artificial Intelligence (AI) is making significant strides in the field of cancer diagnosis in the UK, promising quicker and more accurate identification of various cancers. A study by the Royal Marsden NHS Foundation Trust and the Institute of Cancer Research in London exemplifies this progress, showcasing how AI can be used to grade the aggressiveness of sarcomas, a type of cancer that develops in connective tissues. This AI technology has shown the capability to diagnose and tailor treatment for sarcoma more accurately than traditional biopsies. Researchers focused on retroperitoneal sarcoma, a challenging subtype due to its location at the back of the abdomen. The AI algorithm developed used CT scans from patients, demonstrating an 82% accuracy in grading tumor aggressiveness, significantly higher than the 44% accuracy rate of biopsies. It also distinguished between leiomyosarcoma and liposarcoma in 84% of cases, outperforming radiologists who struggled in 35% of cases.
Medmin has launched a Sarcoma Clinic https://birminghamsarcomaclinic.com which will operate from The Priory and the soon to be opened HCA Harborne Hospital. Surgery for sarcoma is highly specialised and must be performed by a surgeon with appropriate experience in conjunction with specialist pathologists, radiologists, oncologists and nurses, making up a multi-disciplinary team (MDT).
In addition to sarcomas, AI tools are being deployed across NHS hospitals to enhance lung cancer diagnosis. The UK government has allocated £21 million to 64 NHS trusts for the implementation of these AI tools, which analyse X-rays and CT scans to speed up the diagnosis and treatment process. This technology is expected to improve efficiency, reduce waiting times, and assist in managing over 600,000 chest X-rays performed monthly in England.
Despite these advancements, there are challenges hindering the widespread rollout of AI in cancer diagnosis. A key issue is the isolated development of many technological solutions, which may struggle to achieve routine clinical use. This is partly due to limited opportunities for interaction among clinicians, radiologists, scientists, and other experts to collectively understand and address the needs, risks, and challenges in developing, testing, validating, and adopting these tools. The success of AI in cancer diagnosis thus requires the nurturing of multidisciplinary ecosystems, involving commercial partners and stakeholders.
Moreover, the reliance on human expertise in evaluating medical images poses another challenge. Expert radiologists currently set the reference standards for AI and machine learning (ML) evaluations, analysing the presence or absence of disease, tumour boundaries, and responses to treatment. The transition to AI and ML in cancer imaging involves analysing smaller subunits of medical images (pixels/voxels) to discover objective mathematical features that may be linked to disease behaviour or outcomes. This shift requires a careful balance between human expertise and AI capabilities.
In summary, AI’s role in cancer diagnosis in the UK is growing, with promising applications in sarcoma and lung cancer. However, the successful integration of these technologies into routine clinical practice requires overcoming challenges related to interdisciplinary collaboration and the balance between human and AI-driven evaluations.