Exploring Dual-Energy CT Spectral Information for Machine Learning-Driven Lesion Diagnosis in Pre-Log Domain.

Area of Science:

• Medical Imaging
• Radiology
• Computer-Aided Diagnosis

Background:

• Dual-energy spectral CT (DECT) offers advanced imaging capabilities.
• Effective utilization of spectral information for lesion characterization remains a challenge.

Purpose of the Study:

• To propose and evaluate a novel computer-aided diagnosis (CADx) framework, CADxDE, designed for dual-energy spectral CT (DECT).
• To leverage spectral information for improved lesion diagnosis by analyzing material identification and machine learning (ML) based approaches.

Main Methods:

• Developed CADxDE framework operating on pre-log domain transmission data.
• Integrated material identification and ML-based CADx using virtual monoenergetic images (VMIs) generated from decomposed material images.
• Employed an image-driven multi-channel 3D CNN and extracted lesion feature-based ML methods.

Main Results:

• CADxDE achieved higher AUC scores compared to conventional DECT and CT data across three clinical datasets.
• Demonstrated AUC score improvements ranging from 4.01% to 14.25%.
• Mean AUC gain exceeded 9.13%, indicating significant enhancement in lesion diagnosis performance.

Conclusions:

• The CADxDE framework effectively utilizes energy-enhanced tissue features from DECT for improved lesion diagnosis.
• The proposed method shows great potential for enhancing the differentiation of malignant from benign lesions.