Pulmonary Nodule Detection and Classification Using All-Optical Deep Diffractive Neural Network.

Area of Science:

• Optics
• Artificial Intelligence
• Medical Imaging

Background:

• Deep diffractive neural networks (D2NNs) are advanced optical computing structures utilized in image classification and logical operations.
• Computed tomography (CT) imaging is a standard and reliable technique for the detection and analysis of pulmonary nodules.
• Lung cancer diagnosis relies heavily on accurate identification and classification of pulmonary nodules from CT scans.

Purpose of the Study:

• To investigate the efficacy of an all-optical D2NN for the detection and classification of pulmonary nodules in CT images for lung cancer diagnosis.
• To evaluate the performance of the D2NN in identifying the presence of nodules and distinguishing between benign and malignant types.

Main Methods:

• An all-optical deep diffractive neural network (D2NN) was designed and implemented.
• The D2NN was trained using the LIDC-IDRI dataset, a comprehensive collection of lung nodule CT images.
• Performance evaluation was conducted on a separate test set comprising CT images.

Main Results:

• For pulmonary nodule detection, the D2NN achieved a recall rate of 91.08% in identifying the existence of nodules from CT images.
• In pulmonary nodule classification, the D2NN distinguished between benign and malignant nodules with an accuracy of 76.77% and an AUC of 0.8292.

Conclusions:

• The study demonstrates the feasibility of employing optical neural networks for rapid medical image processing.
• All-optical D2NNs offer a potential pathway for developing faster, AI-driven aided diagnostic tools in radiology.
• This approach holds promise for improving the efficiency and accuracy of lung cancer screening and diagnosis.