Automatic recognition of adrenal incidentalomas using a two-stage cascade network: a multicenter study

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• The increasing incidence of adrenal incidentalomas (AIs) necessitates efficient detection methods.
• Early identification of AIs is crucial for managing adrenal diseases like Cushing's syndrome and pheochromocytoma.

Purpose of the Study:

• To develop and validate a deep learning-based automated system for adrenal incidentaloma detection.
• To assess the performance of a two-stage cascade network for AI identification in nonenhanced CT scans.

Main Methods:

• A multicenter retrospective study involving 778 patients.
• Development of a two-stage cascade network comprising a 3D Res-Unet for adrenal gland segmentation and a classifier for AI detection.
• Evaluation using Dice similarity coefficient (DSC), Area Under the Receiver Operating Characteristic Curve (AUC), accuracy, sensitivity, and specificity.

Main Results:

• The model achieved high performance in the validation set with AUCs of 88.15% for left AIs and 87.90% for right AIs.
• No significant difference was found between the deep learning model and manual segmentation (p > 0.05).
• In the test cohort, the cascade network demonstrated AUCs >80% and accuracy >75% for both adrenal glands.

Conclusions:

• A two-stage cascade network utilizing deep learning enables automatic recognition of adrenal incidentalomas.
• This deep learning algorithm is effective for AI detection in nonenhanced CT scans across different medical centers.