Colorizing Grayscale CT images of human lungs using deep learning methods.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Image colorization is a computer-aided rendering technology that transfers colors to grayscale images.
• Deep learning has significantly advanced image colorization techniques in recent years.
• Existing methods often struggle with realistic color transfer for medical imaging.

Purpose of the Study:

• To develop and evaluate deep learning-based methods for colorizing grayscale lung CT images.
• To explore both exemplar and automatic colorization approaches for medical image enhancement.
• To assess the feasibility of using non-medical datasets for training medical image colorization models.

Main Methods:

• Formulated exemplar and automatic image colorization methods.
• Utilized a hybrid approach, selecting reference images for CT scan colorization.
• Trained deep learning models using a dataset of meat images, leveraging their color resemblance to human lungs.
• Extracted pixel features from meat images for automatic colorization of lung CT scans.

Main Results:

• Deep learning models produced significantly genuine and promising colorized lung CT images compared to other methods.
• Image similarity metrics, Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM), achieved satisfactory performance (up to 28.0 and 0.55, respectively).
• The colorization approach demonstrated effectiveness in rendering realistic visual details.

Conclusions:

• Deep learning-based image colorization, trained on analogous color datasets, is effective for lung CT scans.
• The proposed methods offer a promising avenue for enhancing medical image visualization.
• This approach may inspire novel applications for rendering grayscale X-ray images in various security screening contexts.