There is more evidence that artificial intelligence may have a role to play in helping doctors diagnose deadly malignant mesothelioma earlier.
A study conducted at Japan’s Hyogo College of Medicine used past patient records to test a deep convolutional neural network (CDNN). This is a complex computer program designed to evaluate diagnostic criteria.
Pleural mesothelioma is a difficult cancer for doctors to diagnose. Artificial intelligence can help by quickly comparing diagnostic data to hundreds of past cases. In the Japanese study, the AI produced the most accurate mesothelioma diagnoses when combined with all of the available patient data.
The Challenge of Diagnosing Mesothelioma
Mesothelioma tumors grow on the membranes around internal organs. They usually come from past exposure to asbestos. These tumors are thin and irregularly shaped. This makes them hard to see on imaging studies.
Mesothelioma symptoms do not make diagnosis any easier. Patients often complain of vague symptoms like coughing, chest pain, breathing difficulty, and fatigue. These can all be signs of other, less deadly illnesses like pneumonia.
In addition, most doctors see very few cases of mesothelioma in their careers. That is where artificial intelligence has an advantage. AI can “learn” what to look for by analyzing hundreds of mesothelioma cases. It can compare a new case to the previous data in seconds to find a “match”.
Using Artificial Intelligence to Find Mesothelioma
Cancer is not the only disease that can affect the pleural membrane. The pleura can be affected by benign diseases, too. But it is not easy to tell the difference between mesothelioma and a non-cancerous problem.
FDG-PET and CT scans can help with this if doctors know exactly what to look for. Since most people have not looked at hundreds of these scans in mesothelioma patients, they may miss the signs of cancer.
The Japanese scientists wanted to see how well artificial intelligence compared to humans in finding mesothelioma. They used the records of 875 patients with proven or suspected mesothelioma to “teach” the AI what to look for.
Then, they used the records of 174 patients to validate the AI’s programming. Finally, the research team used 176 patient records to compare the diagnostic accuracy of artificial intelligence with other methods. They tried four ways or “protocols” for identifying mesothelioma.
- Protocol A: AI using PET/CT data alone
- Protocol B: a human visual reading of patient scans
- Protocol C: a quantitative method incorporating uptake of the FDG tracer on PET
- Protocol D: AI using PET/CT data, uptake data, gender, and age
“Protocol D showed significantly better diagnostic performance as compared to A, B, and C,” writes lead author Kazuhiro Kitajima.
The research team concluded that artificial intelligence helps tell the difference between mesothelioma and benign pleural diseases. This is important because it means that patients could start the right treatment earlier. Earlier intervention increases the odds of surviving mesothelioma.
“Deep learning with 3D DCNN in combination with FDG-PET/CT imaging results as well as clinical features comprise a novel potential tool [that] shows flexibility for differential diagnosis of MPM,” concludes the report.
Kitajima, K, et al, “Deep learning with deep convolutional neural network using FDG-PET/CT for malignant pleural mesothelioma diagnosis”, June 8, 2021, Oncotarget, PP. 1187 – 1196, https://www.oncotarget.com/article/27979/text/