Two-Stage Deep Learning Model for Adrenal Nodule Detection on CT Images: A Retrospective Study.

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate detection and classification of adrenal nodules are critical for effective patient management.
• Incidental adrenal nodules are frequently discovered during abdominal imaging.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for automated detection and segmentation of adrenal nodules on CT images.
• To assess the DL model's performance in simulating triaging when combined with human interpretation.

Main Methods:

• A retrospective study utilizing internal and external datasets for training and testing a two-stage DL model (detection and segmentation).
• Model performance evaluated using Area Under the Receiver Operating Characteristic Curve (AUC) for detection and Intersection over Union (IoU) for segmentation.
• Simulated triaging performance was assessed by combining DL model output with human interpretation.

Main Results:

• High AUCs for detecting right (0.98) and left (0.93-0.97) adrenal nodules were achieved across internal and external test sets.
• Median IoU values of 0.64 (right) and 0.53 (left) indicated good segmentation performance.
• Combined DL model and human interpretation demonstrated high sensitivity (up to 100%) and specificity (up to 99%), with triaging performance ranging from 0.77 to 0.98.

Conclusions:

• The developed deep learning model exhibits high performance in detecting adrenal nodules.
• This AI tool holds significant potential to improve the detection rates of incidental adrenal nodules.
• The model's integration with human interpretation can enhance diagnostic accuracy and workflow efficiency.