A Physics-Informed Deep Neural Network for Harmonization of CT Images

• Published in: IEEE transactions on biomedical engineering Vol. 71; no. 12; pp. 3494 - 3504
• Main Authors: Zarei, Mojtaba, Sotoudeh-Paima, Saman, McCabe, Cindy, Abadi, Ehsan, Samei, Ehsan
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.12.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Artificial neural networks; Biomarkers; Biomedical imaging; Computational modeling; Computed tomography; Datasets; Deep Learning; Emphysema; Harmonization; Humans; Image acquisition; Image Processing, Computer-Assisted - methods; Image quality; Image reconstruction; Lung - diagnostic imaging; Lung diseases; Lung Diseases - diagnostic imaging; Lung nodules; Lungs; Medical imaging; Neural networks; Neural Networks, Computer; Nodules; Phantoms, Imaging; Physics; physics-informed deep learning model; quantification; Radiographic Image Interpretation, Computer-Assisted - methods; Radiomics; Signal to noise ratio; Tomography, X-Ray Computed - methods; Training; Virtual networks
• ISSN: 0018-9294; 1558-2531; 1558-2531
• DOI: 10.1109/TBME.2024.3428399

Objective : Computed Tomography (CT) quantification is affected by the variability in image acquisition and rendition. This paper aimed to reduce this variability by harmonizing the images utilizing physics-based deep neural networks (DNNs). Methods : An adversarial generative network was trained on virtual CT images acquired under various imaging conditions using a virtual imaging platform with 40 computational patient models. These models featured anthropomorphic lungs with different levels of pulmonary diseases, including nodules and emphysema. Imaging was conducted using a validated CT simulator at two dose levels and varying reconstruction kernels. The trained model was tested on an independent virtual test dataset and two clinical datasets. Results : On the virtual test set, the harmonizer improved the structural similarity index from 79.3 <inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula> 16.4% to 95.8 <inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula> 1.7%, normalized mean squared error from 16.7 <inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula> 9.7% to 9.2 <inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula> 1.7%, and peak signal-to-noise ratio from 27.7 <inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula> 3.7 dB to 32.2 <inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula> 1.6 dB. Moreover, the harmonized images yielded more precise quantification of emphysema-based imaging biomarkers for lung attenuation, LAA −950 from 5.6 <inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula> 8.7% to 0.23 <inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula> 0.16%, Perc 15 from 43.4 <inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula> 45.4 HU to 20.0 <inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula> 7.5 HU, and Lung Mass from 0.3 <inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula> 0.3 g to 0.1 <inline-formula><tex-math notation="LaTeX">\pm</tex-math></inline-formula> 0.2 g. In clinical data, the harmonizer reduced biomarker variability by an average of 70%. For lung nodules, harmonized images improved the detectability index by 6.5-fold and DNN-based precision by 6%. Conclusion : The proposed harmonizer significantly enhances image quality and quantification accuracy in CT imaging. Significance: The study demonstrated the potential utility of image harmonization for consistent CT image quality and reliable quantification, which is crucial for clinical applications and patient management.