| | |

Machine Learning Tool May Predict Mesothelioma Survival Better Than Pathologists

machine learning tool

The developers of a machine learning tool called MesoNet say the program can predict mesothelioma survival better than a pathologist.

MesoNet is a deep convolutional neural network. It analyzes digitized images of mesothelioma cells to predict overall survival.

In a new article in Nature Medicine, its creators compared the machine learning tool with pathologist-read slides. Then they validated MesoNet with two different sets of mesothelioma patients. 

The result may offer a new, more accurate approach to mesothelioma treatment planning

What is a Machine Learning Tool?

A machine learning tool is a computer program that gets “smarter” the more it runs. A deep convolutional neural network is a type of machine learning tool that analyzes images. The more images it analyzes, the better it gets at analysis. 

One challenge in predicting mesothelioma survival is that mesothelioma is very rare. With the exception of mesothelioma specialists, most doctors never even see a single case. 

The same is true for pathologists. If they do not read many mesothelioma slides, they are less likely to detect nuances that may influence survival.

Malignant mesothelioma is a highly complex and heterogeneous disease, rendering its diagnosis and histological typing difficult and leading to suboptimal patient care and decisions regarding treatment modalities,” write the MesoNet testers.

By “training” MesoNet with thousands of mesothelioma cell images, its developers hope to get around the problem of low-volume. 

Mesothelioma Subtypes and Mesothelioma Survival

Malignant mesothelioma cells fall into three major subtypes. The subtypes are based on subtle differences in the cells. Only a pathologist – or a trained machine learning tool – can spot the differences. 

Patients with different subtypes of mesothelioma may have exactly the same symptoms. But they may respond differently to treatment. This can influence their likelihood of survival. 

The most common mesothelioma subtype is epithelioid mesothelioma. This accounts for about 75 percent of cases. Epithelioid cells tend to respond best to mesothelioma therapies. 

Ten to 20 percent of mesothelioma cases fall into the category of sarcomatoid. Sarcomatoid mesothelioma is harder to treat and carries a worse prognosis. Biphasic mesothelioma – the least common subtype – contains some of both types of cells. It is also the hardest to treat.

Predicting Mesothelioma Treatment Response

New tests show the MesoNet machine learning tool recognizes mesothelioma subtypes quickly and accurately. 

Researchers tested the tool using data from two large databases of cancer patients. “The model was more accurate in predicting patient survival than using current pathology practices,” write the researchers. 

The machine learning tool did not just accurately predict mesothelioma survival. It also revealed the basis for those predictions. It turns out that mesothelioma survival may have more to do with things like cellular diversity, inflammation, and vacuolization that doctors thought.

“These findings suggest that deep learning models can identify new features predictive of patient survival and potentially lead to new biomarker discoveries,” the report concludes. 


Courtiol, P, “Deep learning-based classification of mesothelioma improves prediction of patient outcome”, October 7, 2019, Nature Medicine, Epub ahead of print, https://www.nature.com/articles/s41591-019-0583-3

Similar Posts