DCE-MRI-derived peritumoral and habitat radiomics for preoperative prediction of p53-abnormal endometrial cancer: a dual-center retrospective study with external validation
This dual-center study with external validation demonstrates that a dynamic contrast-enhanced MRI fusion model integrating clinical predictors, peritumoral, and intratumoral habitat radiomics features effectively and noninvasively predicts p53-abnormal endometrial cancer, offering a robust tool for preoperative molecular risk stratification.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Endometrial cancer, a disease affecting the lining of the uterus, is becoming more common, and its behavior varies wildly from patient to patient. For decades, doctors relied on broad categories to decide how to treat it, but a modern understanding of genetics has revealed that the disease is actually driven by different molecular engines. One of the most dangerous of these engines involves a gene called p53. When this gene malfunctions, the cancer tends to grow aggressively and resist standard treatments, making it crucial to identify these specific cases before surgery begins. Currently, the only way to know for sure if a patient has this dangerous subtype is to take a tissue sample and examine it under a microscope or run complex genetic tests. However, these methods have flaws. A small tissue sample might miss the dangerous part of the tumor because the cancer is not uniform throughout, much like trying to guess the flavor of a whole cake by tasting just one crumb. Furthermore, the microscopic examination can be subjective, leading to different doctors reaching different conclusions about the same sample.
Because of these limitations, researchers have been searching for a way to see the tumor's true nature without cutting into it. They turned to magnetic resonance imaging, or MRI, which creates detailed pictures of the body's interior. While doctors have long used these images to see the size and shape of a tumor, a newer approach called radiomics allows computers to extract thousands of tiny, invisible details from the picture that the human eye cannot see. This study set out to build a computer model that could act as a non-invasive detective, using these hidden image details to predict whether a patient has the dangerous p53-mutated cancer before they ever step into an operating room.
The researchers, working across two major hospitals, gathered data from nearly 500 women who had undergone surgery for endometrial cancer. They knew the true genetic status of every patient's tumor because each case had been confirmed with advanced genetic sequencing, which served as the gold standard for truth. The team then went back to the preoperative MRI scans these women had received. Instead of just looking at the tumor as a single, solid lump, they broke the image down into three distinct zones of interest. First, they analyzed the tumor itself. Second, they looked at a thin ring of tissue immediately surrounding the tumor, extending five millimeters outward, to see how the cancer was interacting with its immediate neighborhood. Third, and perhaps most innovatively, they used a computer algorithm to divide the inside of the tumor into three different functional zones, or "habitats." This step was designed to capture the fact that the inside of an aggressive tumor is often a chaotic mix of dead cells, dense living cells, and areas with poor blood supply, rather than a uniform block of tissue.
From these three zones, the computer extracted over eleven thousand mathematical features describing the texture, brightness, and patterns of the images. The team then trained a machine-learning system to find which of these thousands of features were most useful for spotting the dangerous p53-mutated cancer. They tested different ways of combining this information, creating separate models that looked only at the tumor, only at the surrounding ring, or only at the internal habitats. They also built a final "fusion" model that combined the best imaging clues with simple clinical facts about the patients, such as their blood pressure and blood counts.
The results showed that looking at the tumor's internal complexity was the single most powerful way to spot the dangerous subtype. The model that analyzed the three internal habitats alone was remarkably stable, correctly identifying the aggressive cancer in about 81 percent of cases in the initial group and maintaining that accuracy when tested on a completely different group of patients from a second hospital. However, the most accurate tool was the fusion model. By combining the deep insights from the tumor's internal habitats and its surrounding ring with the patient's clinical history, the system achieved a high level of discrimination, correctly identifying the dangerous cases in nearly 89 percent of the training cases and 86 percent of the external test cases. Crucially, the model was exceptionally good at ruling out the dangerous cancer; if the model said a patient did not have the p53 mutation, there was a 97 percent chance that was correct.
The study also revealed why this approach works better than previous methods. The researchers found that the dangerous cancer leaves a specific signature not just in the tumor's core, but in how it reshapes the tissue around it and how it creates a chaotic internal environment. While the model that looked only at the tissue surrounding the tumor was less consistent when tested on patients from a different hospital, likely because that area is harder to define precisely on a scan, the internal habitat analysis remained robust. This suggests that the true biological fingerprint of this aggressive cancer is deeply embedded in the tumor's own complex structure. The researchers concluded that this imaging-based approach does not replace the need for genetic testing, but it offers a powerful, non-invasive way to screen patients. It could help doctors prioritize who needs urgent genetic testing and spare low-risk patients from unnecessary invasive procedures, ensuring that the most aggressive cases are identified and treated with the right intensity from the very beginning.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.