The AI doctor will see you now, or at least take a look at your CT-scan data.
Researchers at the Royal Marsden NHS Foundation Trust and the Institute for Cancer Research have developed and tested an AI model that analyses CT scan data of suspected cancer patients. The AI model was shown to be almost twice as good as humans at grading the aggressiveness of a tumour. The AI model also outperformed humans at differentiating between different types of tumour.
The model was trained to characterise retroperitoneal sarcomas, a type of tumour occurring in the lower abdomen, and currently with poor prognosis. An increase in early and accurate diagnosis could significantly improve the potential outcome for patients, while avoiding unnecessary stress and additional tests for people labelled as low-risk.
The study is a great example of AI working alongside humans to improve the patient experience in healthcare.
The article states that “by giving clinicians a more accurate way of grading tumours, researchers hope AI will improve outcomes for patients. Because high-grade tumours can indicate aggressive disease, the new tool could help to ensure those high-risk patients are identified more quickly and treated promptly.”
While the study is limited to retroperitoneal sarcomas, the researchers are hopeful that the methods can be expanded to other types of tumours, and even other diseases.
Unlike generative AI, which has recently been a frequent topic of controversy, the AI model for grading tumours is an example of narrow AI, trained to perform a particular task. In several technology sectors, specific AI applications are starting to reach maturity and finding more and more uses in the real world.
The study aligns well with the findings in the Marks & Clerk 2023 AI report, which highlights the importance of the medtech sector, which shows strong patent filing growth at the European Patent Office.
Nearly 50% of medical AI applications published at the EPO in 2022 relate to either medical imaging or diagnostics, the two largest subcategories within the medtech category. The report notes that many of the publications classified as medical imaging are likely also classified under diagnostics. The sarcoma grading AI model discussed in the article is a perfect illustration of this overlap in classifications, where image analysis is used for diagnosis.
The success of the AI model grading is a promising sign that AI applications may be able to take some of the load on overstretched health services, and simultaneously improve patient prognosis by delivering an early and more accurate diagnosis.