An AI model developed by researchers at the Antoni van Leeuwenhoek (AVL) in the Netherlands and the Dutch Cancer Institute (NKI) can determine more accurately whether a treatment for pleural cancer is effective than existing measurement methods. The model, called ARTIMES, analyses the entire tumour volume on CT scans, thereby providing doctors with a more reliable tool for making treatment decisions. According to the researchers, the technology could not only improve care for patients with mesothelioma, but may also transform the assessment of other tumour types.
Assessing the effectiveness of cancer treatment is usually done by measuring whether a tumour is growing or shrinking. The so-called RECIST criteria, which are based on measuring tumour cross-sections on scans, are used worldwide for this purpose. However, in the case of mesothelioma, an aggressive form of pleural cancer, this method falls short. The tumour does not grow as a clearly defined mass, but often forms a thin and irregular layer along the lung wall. This makes it difficult to determine where and how measurements should be taken. This regularly leads to uncertainty among both doctors and patients. Furthermore, it can take longer before it becomes clear whether a treatment is effective.
AI measures tumour volume
To solve this problem, AI specialists, radiologists and pulmonologists at the AVL developed the AI model ARTIMES. Instead of measuring just a few dimensions, the model analyses the entire volume of the tumour and compares it with previous scans. According to pulmonologist Sjaak Burgers, such an analysis is virtually impossible for humans to carry out. Manually assessing every single pixel on a CT scan would be extremely time-consuming. AI, on the other hand, can carry out this task quickly and consistently, after which doctors check and interpret the results.
More than 11,000 CT scans from over 2,000 patients across 121 hospitals worldwide were used to develop and validate the model. This makes the study one of the largest in this field. Medical technologist Kevin Groot Lipman, who led the research, states that ARTIMES demonstrates for the first time that AI outperforms human assessment alone in this application and that doctors can actually use the results in clinical decision-making. The results of the study have been published in the scientific journal The Lancet.
Faster course correction
The model’s added value lies primarily in its ability to identify treatments that are ineffective more quickly. According to the researchers, when a therapy is not working, this can be detected earlier than with current measurement methods. This offers patients the opportunity to switch to an alternative treatment more quickly. At the same time, unnecessary side effects from ineffective therapies can be prevented and healthcare costs can be reduced.
To ensure the technology is immediately applicable in clinical practice, the research team developed not only the AI model but also guidelines to support doctors in interpreting the results. The ultimate responsibility for treatment decisions always remains with the doctor.
A new standard?
In addition to its direct impact on patient care, the researchers expect ARTIMES to influence the development of new cancer therapies. Analyses of data from eight clinical trials showed that volume measurements using AI predict how patients fare in the long term more accurately than the RECIST criteria currently in use. As a result, clinical trials could potentially be conducted more efficiently, and it may be possible to determine more quickly whether new treatments are actually effective.
For the time being, due to European regulations, the model may only be used clinically within the Antoni van Leeuwenhoek Hospital. The research team is working towards broader certification so that other hospitals can also make use of the technology. Meanwhile, similar AI models are already being developed at the AVL for conditions including lung cancer and brain metastases. According to the researchers, ARTIMES may mark the beginning of a new generation of oncological measurement methods, in which the focus is no longer on the size of a tumour, but on the total tumour volume.
International research
Last year, an international study led by the Icahn School of Medicine at Mount Sinai already demonstrated that AI can significantly speed up the diagnosis of lung cancer. The AI model developed can predict genetic EGFR mutations, which are important for targeted cancer treatments, based on standard H&E-stained pathology images. As a result, costly and time-consuming genetic tests are not always immediately necessary.
That model was trained using the largest available dataset of lung adenocarcinoma patients from Europe and the United States. In a real-time ‘silent’ clinical trial at the Memorial Sloan Kettering Cancer Centre, the AI accurately detected EGFR mutations and could reduce the number of urgent genetic tests by more than 40 per cent.
