MRI Radiomics Meets Artificial Intelligence: Development and Validation of a Random Forest Model for Brain Tumor Classification in Sudanese Patients

The Invisible Clues in a Brain Scan

Imagine you are a detective trying to solve a mystery, but the only evidence you have is a black-and-white photograph. Sometimes, the photo shows a clear, harmless lump; other times, it shows a dangerous, growing monster. The problem is, to the naked eye, these two things can look surprisingly similar. This is the daily challenge for doctors trying to figure out if a brain tumor is "benign" (harmless) or "malignant" (dangerous). They rely on MRI scans, which are like super-powered cameras that take pictures of the inside of your head using magnets instead of light. But human eyes can get tired, and sometimes two doctors might look at the same picture and disagree on what they see.

To fix this, scientists have invented a new kind of detective work called Radiomics. Think of a standard MRI scan as a painting. A radiologist looks at the painting and says, "That looks like a tumor." Radiomics, however, is like taking that painting and breaking it down into millions of tiny pixels, then measuring the exact shade of gray, the texture, and the shape of every single one. It turns the image into a massive spreadsheet of numbers that the human eye can't see, but a computer can read. By feeding these numbers into Artificial Intelligence (AI)—specifically a type of computer brain called a Random Forest—scientists hope to teach the computer to spot the hidden patterns that tell the difference between a harmless lump and a dangerous one, without needing to cut into the patient. This paper asks a simple but vital question: Can this digital detective work help doctors in Sudan make better decisions about brain tumors?

The Digital Detective in Sudan

In this study, a team of researchers from Alzaiem Alazhari University in Sudan decided to build their own digital detective. They wanted to see if they could create a computer model that helps Sudanese doctors tell the difference between benign and malignant brain tumors just by looking at MRI scans. They gathered a group of 120 patients who had already been diagnosed with brain tumors. Half of them (60 patients) had benign tumors, and the other half (60 patients) had malignant ones. These were real cases, confirmed by looking at the tissue under a microscope, so the researchers knew for sure which group was which.

First, the team had to get the data ready. They took the MRI images and cleaned them up, making sure the colors and sizes were consistent, kind of like adjusting the brightness and contrast on a photo so everything looks the same. Then, a skilled radiologist manually drew a line around the tumor on each image to tell the computer exactly where the trouble spot was. Once the tumors were isolated, they used a special computer program to pull out over 100 different "clues" from each tumor. These clues included things like how bumpy the tumor's surface was, how the gray pixels were arranged, and how intense the colors were.

From that huge list of clues, the team hand-picked the six best ones—the most reliable suspects—to use in their model. They then taught a Random Forest algorithm (a type of machine learning that works like a group of experts voting on a decision) how to use these six clues to guess if a tumor was benign or malignant. They split their data into two groups: a "training" group to teach the computer, and a "testing" group to see if the computer actually learned anything.

What the Computer Found

When the computer took its final exam on the testing group, it got it right 71% of the time. That means if you showed the model 100 new brain scans, it would correctly guess the type of tumor for about 71 of them. It was particularly good at spotting the dangerous ones, catching 75% of the malignant tumors (this is called sensitivity). However, it wasn't perfect; it missed some malignant cases and sometimes got confused about benign ones, resulting in a specificity of 66.7%. The overall score, known as the F1-score, was 0.72, and the area under the curve (AUC), which measures how well the model separates the two groups, was 0.764.

The researchers also looked at the patients' backgrounds to see if things like age or gender helped predict the tumor type. They found that age and sex didn't really matter; the computer didn't need to know if the patient was a 20-year-old male or a 60-year-old female to make a good guess. However, they did find something interesting about the size of the tumor. Patients who had larger tumors were much more likely to have neurological symptoms (like headaches or weakness) than those with smaller tumors. The math showed this link was very strong, with a p-value of less than 0.01.

The Verdict and the Caveats

The study concludes that using MRI radiomics with a Random Forest model is a feasible way to help classify brain tumors in Sudan. It suggests that this approach could be a useful, non-invasive tool for doctors who might not have access to the most advanced equipment, offering a second opinion based on numbers rather than just a glance. The model suggests it can support clinical decision-making and improve diagnostic accuracy.

However, the authors are careful not to call this a finished solution. They point out that their group of 120 patients is relatively small, and because they only looked at data from one hospital in the past, the model might not work exactly the same way for everyone else. They also noted that the computer still made mistakes, missing some malignant tumors, which is a serious issue in medicine. The researchers suggest that before this tool can be used widely in hospitals, it needs to be tested on much larger groups of people from different places. They also recommend adding more types of MRI scans and using even more advanced computer methods in the future. For now, this study is a promising first step, showing that the "digital detective" has potential, but it still needs more training before it can solve the case on its own.