Adrenal masses are often overlooked during routine abdominal Computed Tomography (CT) scans. Since these lesions can be malignant or hormonally functional, missed diagnoses can delay necessary treatment. The challenge lies in the small size and anatomical variability of the adrenal glands. A fully automated deep learning adrenal mass model now offers a solution for reliable detection and measurement of these critical lesions.
A recent study developed a U-Net-based deep learning (DL) model. Researchers trained this model using 415 contrast-enhanced CT scans. The study ensured real-world relevance by including an external test set with an incidence rate of 4.8%. The DL model demonstrated high diagnostic performance in mass detection. Furthermore, its overall accuracy reached 96.6% in the external test set. This is a very significant finding.
The model also accurately measured the size of the masses. The predicted diameters showed a strong correlation (ICC: 0.848-0.855) with sizes measured on CT and in surgical specimens. For example, the model successfully identified 44 out of 50 adrenal masses in one test set. This model greatly reduces the subjective impact of manual linear assessment in CT examinations.
Clinical Impact of Deep Learning Adrenal Mass Detection
Adrenal masses, frequently discovered incidentally, pose a diagnostic and management challenge. Approximately 4% to 6% of abdominal CT scans show these incidental lesions. Though most lesions are benign, early and accurate distinction is vital for patient safety. Consequently, clinicians must quickly identify malignant or functional tumors. Conventional radiological interpretation can be time-consuming and prone to overlooking small masses.
This new DL tool offers valuable support to radiologists. The technology accurately detects masses often missed by human readers. Moreover, it provides a precise size estimate, which aids in differential diagnosis. Therefore, integrating this model into clinical workflows promises to improve screening, follow-up, and preoperative evaluations. The combination of DL and human expertise can potentially reduce radiologist workload while minimizing false-positive and false-negative rates. Ultimately, this automated deep learning model presents a critical advancement in imaging technology. The model’s high accuracy for both detection and measurement holds the potential to significantly enhance patient management.
Frequently Asked Questions
Q1: Why are adrenal masses often missed on CT scans?
Adrenal masses are frequently missed because they are small retroperitoneal organs. They occupy less than 1% of an abdominal CT slice. Their shape, size, and location also vary greatly among patients. These factors make them difficult to detect manually.
Q2: How accurate is the new deep learning model for adrenal mass detection?
The proposed deep learning model achieved an overall accuracy of 96.6% in a real-world external test set. Furthermore, its predicted mass diameters showed a high correlation (ICC 0.848-0.855) with both CT-measured and pathologically measured sizes, demonstrating high performance for both detection and measurement.
References
- Kim TM et al. Automated deep learning for detection and measurement of adrenal masses in contrast-enhanced abdominal CT. Eur Radiol. 2026 Jan 29. doi: 10.1007/s00330-025-12314-5. PMID: 41609765.
- Artificial intelligence in adrenal imaging: A critical review of current applications. researchgate.net.
- Using deep learning to detect adrenal lesions in CT scans. digital.nhs.uk.
- Artificial intelligence with a deep learning network for the quantification and distinction of functional adrenal tumors based o – AME Publishing Company. amegroups.cn.
- Two-Stage Deep Learning Model for Adrenal Nodule Detection on CT Images: A Retrospective Study. rsna.org.
- Artificial intelligence and radiomics applications in adrenal lesions: a systematic review. nih.gov.