Deep learning technology showed strong potential in improving ultrasound-based detection of endometrial cancer (EC) in postmenopausal women, offering patients earlier and more reliable diagnoses, according to a study published recently in Frontiers in Oncology.
Researchers reported that a newly developed computer model significantly enhanced accuracy in identifying cancer compared with standard evaluation alone. The system was designed to analyze gynecological ultrasound scans by measuring features such as endometrial thickness, tumor uniformity, and blood flow patterns.
Specialists’ diagnoses served as the benchmark for assessing the model’s accuracy. Researchers said the tool could help frontline doctors, especially in primary care, recognize cancerous changes sooner, ensuring patients receive treatment without delay.
“The DL [deep learning] model demonstrated high accuracy and robustness, significantly enhancing the ability to diagnostic assistance for EC through ultrasound in postmenopausal women,” explained the authors of this study. They continued, “This provides substantial clinical value, especially by enabling less experienced physicians in primary care settings to effectively detect EC lesions, ensuring that patients receive timely diagnosis and treatment.”
Read more about signs and symptoms of EC
This study involved 877 patients with confirmed EC at the Hospital of Traditional Chinese Medicine of Qiqihar between January 2020 and December 2024. A total of 877 ultrasound images were used, divided into 614 for training, 175 for validation, and 88 for testing. The model’s accuracy was measured using the area under the receiver operating characteristic curve (AUC). Results showed an AUC of 0.844 in the training set, 0.811 in the validation set, and 0.858 in the test set.
To reach those results, the system assessed tumor appearance and internal blood flow. Blood flow indicators included resistance index, peak systolic velocity, end-diastolic velocity, and blood flow area. These values reflect how much blood moves through and around the tumor, which can indicate whether growth is likely cancerous.
Researchers concluded that the deep learning model was both accurate and robust, offering significant clinical value. They noted that its use could be especially important for doctors who are less experienced in reading gynecological ultrasounds, allowing them to detect cancerous lesions more effectively.
For patients, these results may translate into faster and more dependable diagnoses, less reliance on specialized centers, and greater confidence in care received at local clinics. Earlier detection of EC could lead to better treatment outcomes and potentially save lives, particularly for postmenopausal women at higher risk.
Sign up here to get the latest news, perspectives, and information about EC sent directly to your inbox. Registration is free and only takes a minute.
