AI’s emerging role in cancer diagnosis

The uses for AI and machine learning in medicine continues to grow, with more recent evidence that it can be a powerful tool helping diagnose disease and improve treatment.


A study conducted by the Royal Marsden Hospital and the Institute of Cancer Research used AI in combination with a technique called radiomics to diagnose a rare cancer more accurately and earlier than ever before.

Radiomics extracts quantitative data from medical images such as MRI or CT scans. Instead of relying on medical professionals to analyse these images, radiomics analyses quantitative data extracted from scans. This is where AI comes in, as it can analyse the massive amounts of data generated from these images, look for patterns, and predict tumours more accurately.

The type of cancer involved in this study was retroperitoneal sarcoma, which are cancerous tumours that develop in the connective tissue at the back of the abdomen, next to the kidneys. Approximately 4,300 people in England are diagnosed with this type of cancer each year, and its diagnosis is notoriously difficult, with under-grading of tumours being common. Currently, doctors will read a CT scan before deciding whether a tissue biopsy is required. An important advantage of this new AI/radiomics-related method is that a more accurate diagnosis can occur directly from the CT scan without the need for an invasive biopsy and a subsequent wait for lab analysis.

In this study, the AI model graded the aggressiveness of tumours from CT scans with 82% accuracy, compared to the 44% accuracy achieved by a clinician reading the scan, taking a tissue sample, and waiting for the results of a lab analysis. Not only is the new method nearly twice as accurate, it is also quicker and spares the patient an invasive biopsy. While accuracy of diagnosis of this rare cancer is extremely promising in itself, the prospect of this combination of radiomics and AI being applied to the diagnosis of a wide array of other cancers has the potential to revolutionise cancer diagnosis and treatment.

This breakthrough is another example of non-generative AI being used in medicine as a tool to deliver a higher standard of healthcare than ever before and coincides with a new breed of data driven organisations (“TechBios”) emerging with the aim of commercialising this type of technology (for more on this topic, see our key takeaways from the TechBio UK 2023 conference here). With UK IPO guidance last year stating that AI inventions can be patentable if certain criteria are met*, and with the MHRA currently preparing detailed guidance on the regulation of AI as a medical device (in addition to their Regulatory Roadmap), the widespread adoption of AI in healthcare continues apace.

*Editor’s note: this guidance has been temporarily suspended pending consideration of the High Court decision in Emotional Perception AI Ltd v Comptroller-General of Patents, Designs and Trade Marks [2023] EWHC 2948 (ch), which was handed down on 21 November 2023 and held that an invention relating to a trained artificial neural network was patentable.