Estimating TP53 Mutations in Endometrial Cancer Using Deep Learning and Radiomics

Endometrial cancer is one of the most common malignancies of the female reproductive system. The presence of TP53 mutations has been linked to more aggressive tumor behavior, resistance to treatment, and poorer survival outcomes. Traditionally, TP53 status is assessed using immunohistochemistry, but this method requires tumor samples, making it invasive and often performed only after surgery. Developing a noninvasive approach to identify TP53 mutations could provide significant clinical benefits by enabling early risk assessment and treatment planning.

Diffusion-weighted imaging is an advanced MRI technique that allows for the analysis of tumor microstructure. Radiomics and deep learning have emerged as powerful tools for extracting detailed imaging features that may not be visible to the human eye. This study aimed to develop a prediction model that combines clinical variables, radiomics, and deep learning features to noninvasively estimate TP53 mutation status in endometrial cancer patients.

How the Study Was Conducted

The study analyzed MRI scans and clinical data from 155 patients with endometrial cancer. Patients were divided into a training set, a test set, and an external validation set. Diffusion-weighted imaging scans were processed to extract radiomic features, and deep learning models were applied to analyze imaging patterns associated with TP53 mutations. Clinical factors, including age, lesion size, and tumor markers, were also included in the analysis.

Feature selection was performed using statistical methods to identify the most relevant predictors. Prediction models were then developed using machine learning algorithms, including Gaussian process and decision tree classifiers. Model performance was evaluated based on diagnostic accuracy, calibration curves, and clinical decision analysis.

Key Findings on TP53 Mutation Prediction

The study found that combining deep learning, radiomics, and clinical variables significantly improved the accuracy of TP53 mutation prediction. Compared to models based on deep learning, radiomics, or clinical features alone, the combined model demonstrated the highest diagnostic performance across all datasets.

The final model, built using the Gaussian process algorithm, included four deep learning features, five radiomics features, and two clinical variables. It achieved an area under the curve of 0.949 in the training set, 0.877 in the test set, and 0.914 in the external validation set, indicating strong predictive capability.

This study highlights the potential of diffusion-weighted imaging, deep learning, and radiomics as noninvasive tools for identifying TP53 mutations in endometrial cancer. The ability to predict TP53 status without the need for a biopsy could help doctors personalize treatment strategies and identify high-risk patients earlier.

Patients with TP53 mutations tend to have more aggressive tumors and may benefit from closer monitoring and tailored therapeutic approaches. The integration of this predictive model into clinical workflows could improve risk stratification, guide treatment decisions, and reduce the reliance on invasive testing methods.

To learn more, read this!: Estimation of TP53 mutations for endometrial cancer based on diffusion-weighted imaging deep learning and radiomics features | BMC Cancer | Full Text