Convolutional Neural Network-Based Diagnostic Model for a Solid, Indeterminate Solitary Pulmonary Nodule or Mass on...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonology

Background:

• Solid indeterminate solitary pulmonary nodules (SPNs) or masses (SPMs) on computed tomography (CT) present a diagnostic challenge.
• Accurate differentiation between benign and malignant lesions is crucial for appropriate patient management.

Purpose of the Study:

• To develop and evaluate a non-invasive diagnostic model using convolutional neural networks (CNNs).
• The model aims to distinguish benign from malignant solid indeterminate SPNs/SPMs on CT scans.

Main Methods:

• A retrospective study included 459 patients with solid indeterminate SPNs/SPMs on CT.
• An end-to-end CNN model was developed and validated on train, validation, and test sets.
• Radiologists' diagnostic performance with and without the CNN model was compared.

Main Results:

• The CNN model achieved an area under the curve (AUC) of 91% in the test set.
• Diagnostic accuracy significantly improved for radiologists using the CNN model (e.g., 85% vs. 66% in the test set).
• Specificity saw a substantial average improvement of 42% with the CNN model.

Conclusions:

• The developed CNN model shows promise as a valuable tool for non-invasive differentiation of benign and malignant SPNs/SPMs on CT.
• This AI-assisted approach can enhance diagnostic precision in pulmonary nodule assessment.