Double paths network with residual information distillation for improving lung CT image super resolution

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate medical image analysis is crucial for disease diagnosis.
• The COVID-19 pandemic highlighted the need for precise diagnostic tools.
• High-resolution lung CT images offer vital diagnostic information, necessitating super-resolution techniques.

Purpose of the Study:

• To develop an advanced super-resolution method for medical images.
• To improve the resolution of lung CT scans for better COVID-19 diagnosis.
• To introduce the Double Paths with Residual Information Distillation for Super Resolution (DRIDSR) network.

Main Methods:

• A novel DRIDSR network was designed, featuring parallel low-frequency and high-frequency paths.
• The low-frequency path utilizes a shallow convolutional network.
• The high-frequency path employs a Residual Information Distillation Module (RIDM) with cascaded residual blocks and information distillation blocks (IDBs).

Main Results:

• DRIDSR demonstrated superior reconstruction quality on the COVID-CT dataset compared to SRCNN, ESPCN, VDSR, IMDN, and PAN methods at ×3 and ×4 upscale factors.
• The method achieved significant PSNR improvements, ranging from +0.43 dB to +2.41 dB.
• DRIDSR also reduced the number of parameters and analysis time.

Conclusions:

• The DRIDSR network achieves superior performance in generating high-resolution medical images.
• It outperforms several state-of-the-art super-resolution methods based on objective metrics and subjective evaluations.
• DRIDSR offers a promising solution for enhancing medical image quality in diagnostic applications.