AI: The Future of Bladder Cancer Diagnostics and Treatment

Bladder cancer is the most expensive cancer to treat over a lifetime. But artificial intelligence (AI) methods may reduce the cost of its diagnosis and treatment. AI can:¹

• Detect tumors
• Identify cancer stages
• Predict response to treatment

This may improve treatment efficiency and health outcomes. Using AI can lead to:^1,2

• Fewer invasive procedures
• Shorter hospital stays
• Fewer complications

But doctors do not yet use AI to diagnose or treat bladder cancer. Several challenges must be overcome to understand how to better use AI methods.²

What is artificial intelligence?

AI is a machine that uses skills linked to human intelligence. In AI, a computer model is programmed by a human to act or use reason to perform a certain task. AI is part of our everyday lives. AI applications include:^1,3

• Virtual assistants on smartphones
• Facial and speech recognition technology
• Spam filters on email inboxes
• Product recommendation systems
• Search engine algorithms
• Self-driving cars
• Generative or creative tools (such as ChatGPT)

What is machine learning?

Machine learning is a part of AI that continually learns. It is not programmed just to perform a specific task. Instead, it learns from data to make predictions or decisions. In this way, machine learning models perform better over time with more data.^3,4

In medicine, data can come from many sources like:¹

• Imaging, such as X-ray and magnetic resonance imaging (MRI) scans
• Wearable sensors
• Electronic health records

In machine learning, computers are trained using this data. Computers then use this data to make decisions or predictions. The series of steps they use to make these predictions are called algorithms. The computers then continue to learn based on new data.¹

How can AI diagnose and treat bladder cancer?

Experts are most excited about using AI to diagnose bladder cancer from medical imaging. AI algorithms may do this as well as or better than doctors. Below are some tasks AI may perform.^1,2,4

Automatic tumor detection

A common way to look for tumors is through a procedure called cystoscopy. But diagnosis from imaging during cystoscopy has a high error rate. AI models can avoid human error to improve accuracy during cystoscopy.^1,5

Several AI models have been developed to do this. Researchers trained the models with images of normal bladder tissue and bladder tumors. These models have over 90 percent accuracy in finding tumors.⁶

Other models can detect bladder tumors from urine samples. Getting urine samples is cheaper and easier than getting images from cystoscopy. But these AI models seem to be less accurate.^7,8

Bladder segmentation

Bladders vary in shape and size. This can make it hard to identify them in medical images. Bladder segmentation is the process of trying to distinguish the bladder from surrounding areas. This helps identify the location of any tumors. Several AI models can find the borders of the bladder on imaging tests like computerized tomography (CT) or MRI scans.^9,10

Tumor staging and grading

Bladder cancer treatment depends on many factors, including the tumor stage. Tumor stage refers to how far the tumor has spread. Its depth within the bladder is especially important. Several AI models can tell tumor stages from CT or MRI scans.^11,12

Treatment also depends on the tumor grade. Tumor grade is based on what the tumor cells look like. Higher-grade tumors are more aggressive and may spread faster. But grading tumors is very subjective because it is based on visual observation. Several AI models improve accuracy by grading tumors from biopsy samples or MRI scans.^13,14

Predicting treatment response and survival

Predicting how well people will respond to chemotherapy can improve treatment outcomes. Some people endure serious side effects without any benefits. Other people respond well and can avoid surgery. Several AI models use data from CT scans to accurately predict what a person's response to chemotherapy will be.^15-17

Survival rate varies based on many factors. Estimating survival rate can help doctors plan treatments. Several AI-based methods can estimate survival rate based on:^1,18

• Gender
• Age
• Tumor stage and grade
• Surgical approach
• Other factors

What challenges must be overcome for AI to be used more often?

Despite promising results from many studies, AI algorithms are not yet a part of diagnosis and treatment for bladder cancer. Major challenges must be addressed for these models to be adopted.^1,2

A major issue is that AI models work better with training data than with real-world data. This may be because of differences in how data is gathered. AI models should be trained on data from different clinics. This would help standardize AI models so they can be used in different kinds of settings.^1,2

Another major issue is trust in AI. Understanding how AI models reach their decisions can be hard. Doctors may not trust models when it is unclear how they work. More research is needed to improve understanding and trust of AI in medical settings. Larger clinical studies of AI models would also help improve the quality of and trust in AI.¹