Explainable AI for Lung Cancer

A new AI-based system provides precise and understandable way to analyze tissue samples from lung cancer patients, potentially leading to more accurate diagnoses and better predictions of patient outcomes.

 

The Growing Challenge of Lung Cancer

In the era of modern medicine, lung cancer remains a major challenge to human health. In fact, out of all cancers, lung cancer is the most diagnosed and has the highest mortality rates. More worryingly, the global burden is expected to rise by 77% by 2050.

This highlights a pressing need for new ways to diagnose and treat this deadly disease.

To help patients, many doctors tailor treatments to each person’s specific needs. This starts with figuring out the exact type of lung cancer a patient has.

The Promise and Pitfalls of AI

Although digitized pathology has been an extremely useful tool, there is growing interest in implementing AI into image analysis.

However, most AI studies in this area aren’t ready for real-world use. One big issue is that some AI methods are like “black boxes” — they’re so complex that even experts can’t fully understand how they make decisions. This lack of transparency can lead to problems with trust and reliability.

 

The Study: Making AI Explainable

Researchers at the University of Cologne (Cologne, Germany) are working on a solution to this problem. Carina Kludt and her team have developed a new method to make AI models more understandable.

They’ve created a fully automated tool for classifying lung cancer patients. The algorithm can identify and separate different types of tissues in images with high accuracy, helping doctors determine the exact type of lung cancer a patient has.

To make sure their AI tool is accurate, the team used a large, diverse set of data from multiple sources. They tested the tool by comparing its decisions to those of experienced doctors. They found that most ‘mistakes’ were due to the tumors lacking clear features, not because the AI was faulty.

 

Predicting Lung Cancer Progression & New Metrics

In addition to improving diagnosis, this AI platform can also predict how a patient’s cancer might progress.

The team created new, easy-to-understand metrics based on the presence of specific tissue features, such as the amount of dead tissue in the tumor and the presence of immune structures known as tertiary lymphoid structures.

These metrics help doctors assess the aggressiveness of the cancer and predict how likely it is to spread or recur, allowing for more personalized and effective treatment plans.

 

A Step Closer to Clinical Implementation

This study marks a significant step towards using AI in clinical settings.  What’s even more important is that the algorithm’s transparency allows for better collaboration between doctors and digital tools.

The versatility of this AI system also means it can be used in many different ways beyond diagnosis and prognosis. For example, it could help doctors decide which patients would benefit most from additional therapies after surgery.

The research team is also making their data and tools available to other scientists, encouraging further innovation and collaboration in the fight against lung cancer.