Artificial intelligence (AI) predicts the best treatment for patients with melanoma. The AI model was developed as part of the Data Science Experience (DSX) project set up by Isala Hospital.
"It is my desire to improve patient care by using available data," says Richard Brohet, genetic epidemiologist, statistician and data scientist at Isala Academy. Brohet, together with colleagues, started the DSX project.
"I want to do that initially for oncology patients. DSX uses all medical data in Isala, or Big Data. We discussed with doctors what kind of algorithm with AI (artificial intelligence) we wanted to develop first. The choice fell on melanoma. Together with internist-oncologist Jan Willem de Groot, I therefore started this pilot study."
What is the best treatment?
Patients with cancer face several issues. Why did it arise and what is the best treatment? "For oncology patients, these questions are difficult to answer. Cancer can arise from various causes, think lifestyle, hormones, environment, genetic disorders, heredity, or a combination of these factors. Also, cancer treatment depends on personal, clinical and genetic characteristics. Because every tumour and therefore every patient is unique, we want to tailor treatment so that survival is better and the chance of treatment not working is lower. Patients with melanoma now receive targeted therapy or immunotherapy in addition to chemotherapy. This works in some patients, but not in others. How is this possible? AI could help enormously in answering this question and delivering tailor-made treatment. So that is what we are now trying to do with DSX," Brohet explains.
The AI model developed as part of the DSX project deploys structured and unstructured data for analysis. Structured data include gender, age, weight and tumour size. Unstructured data includes data entered in open text fields. These include how a person is doing, their lifestyle and occupation, and how the patient is coping with the disease. This is specifically information that does not fit into a checklist, but is important.
Seeing connections
De Groot: "By also using that data, you can look for what you don't expect and see correlations. It also gives us insight into the outcomes of different treatments over time. We can capture and analyse more and more complex data this way and we can better see what predictive factors are."
The basic AI model that has now been developed predicts the best treatment for a patient with melanoma. However, the model needs further validation. De Groot reports that if he inputs data from a known patient, the model does indeed predict the correct treatment. The algorithm now needs to grow, which the creators want to do by adding more patients, including from other melanoma centres. In this way, they want to further optimise the predictions and thus offer customised treatment to patients. De Groot: "People often look for drugs that can stretch life by a few months. My opinion is that we are better off investing in quality of life. We can do this by learning from the available data to decide on the most appropriate treatment together with the patient."
Determining risk of lung cancer
Researchers have been working for some time on AI models to support the treatment of cancer patients. For example, an AI algorithm developed by researchers at Radboudumc assesses lung nodules. The model predicts the risk of these nodules developing into tumours. Detecting tumours at an early stage increases the chances of successful treatment and reduces the risk of death from lung cancer. However, many of the small nodules visible on CT scans are benign. These can be left alone, whilst malignant nodules do need to be treated. The algorithm distinguishes between benign and malignant nodules.
A second example is an algorithm developed by researchers at Eindhoven University of Technology. This is a machine learning model that predicts whether immunotherapy will be effective for a cancer patient. One of the roles of our immune system is to eliminate threats within our body, including cancer. However, cancer cells can disable immune cells, thereby preventing their own destruction. Immunotherapy can reverse this process in some cases. The algorithm predicts the likelihood of successful treatment with immunotherapy.
Fighting uterine cancer
Through the PRESCRIP-TEC research project, in which the LUMC is one of the participating organisations, researchers are working on AI to help combat cervical cancer. This form of cancer is caused by the human papillomavirus (HPV). Women with a chronic HPV infection develop precancerous lesions of cervical cancer. These can be detected through screening. In the Netherlands, screening via smear tests is commonplace. In many low- and middle-income countries, however, access to healthcare is often poor and screening is rare.
A self-test can help, after which a relatively simple treatment is possible. However, these are relatively labour-intensive in remote areas, while rapid intervention is important. Through the PRESCRIP-TEC project, researchers are working on AI that helps assess self-tests that women perform on themselves with a cotton swab. If they have HPV and require screening, AI technology helps health workers recognise the discolouration of the cervix.
Author: Wouter Hoeffnagel
Photo: Gordon Johnson via Pixabay
