Anatomically adaptive feature-wise linear modulation for deep learning-based low-dose CT denoising

Clinical Snapshot

Key Findings

• P Value: Not reported

• Effect Size: Noise reduction percentages of 35–49% across tissue types; no standardised effect size (e.g., Cohen's d) or absolute image quality metric values (PSNR, SSIM) reported in the abstract

• Primary Outcome: Tissue-stratified noise reduction: 37.14% in lung, 48.75% in soft tissue, and 35.18% in bone regions on the AAPM-Mayo Clinic LDCT dataset. The model is reported to outperform all competing deep learning denoising methods evaluated.

• Nnt Or Sensitivity: Not applicable in conventional clinical sense. No diagnostic sensitivity/specificity, AUC, or NNT data provided. The relevant technical analogue — improvement in PSNR/SSIM over comparators — is not quantified in the abstract.

• Confidence Interval: Not reported

Clinical Application

Clinical deployment would require integration into PACS/CT reconstruction pipelines, prospective validation on local scanner hardware, and demonstration of diagnostic non-inferiority compared to standard-dose CT. Computational requirements for real-time inference are not described. Regulatory approval (TGA in Australia, FDA in the US, CE marking in Europe) would be mandatory prior to clinical use. The model weights and code availability are not mentioned, limiting reproducibility. Australia's emerging national lung cancer screening program and existing high-volume CT infrastructure make LDCT denoising clinically relevant. The TGA regulates AI/ML-based medical imaging software as a Class IIa or higher medical device under the Therapeutic Goods (Medical Devices) Regulations 2002, requiring conformity assessment before clinical use. The RACGP and RANZCR have not issued specific guidance on AI-based CT denoising at the time of this appraisal. PBS does not currently fund AI-assisted image processing as a standalone item. Australian CT practice predominantly uses GE, Siemens, and Canon scanners — validation on these platforms would be required. The AAPM-Mayo dataset was acquired on Siemens scanners, providing partial but incomplete relevance to the Australian scanner mix. Patients undergoing LDCT examinations where radiation dose reduction is clinically indicated — including lung cancer screening, paediatric CT, serial imaging in oncology follow-up, and CT pulmonary angiography in younger patients. The framework is potentially applicable across thoracic, abdominal, and musculoskeletal LDCT protocols given the multi-tissue design.

Abstract

Objective.Low-dose computed tomography (LDCT) reduces radiation dose but, introduces heterogeneous noise due to different photon attenuation based on anatomical tissue. Most deep learning techniques assume uniform noise in LDCT and perform equal noise removal across different regions, leading to sub-optimal performance across different tissues. This work aims to design a physics-based framework that explicitly models region-dependent noise characteristics to improve LDCT noise removal.Approach.We propose an anatomically adaptive noise reduction framework. The proposed anatomically adaptive feature-wise linear modulation (FiLM) model consists of a U-Net architecture that integrates two complementary units: the global FiLM unit in the encoder, which modifies features globally based on global image statistics to remove overall noise, and the local FiLM unit in the decoder, which modifies features locally based on the type of anatomical tissue. This dual design enables the modeling of overall image noise characteristics in addition to the removal of local noise associated with each anatomical tissue.Main results.The model performance was evaluated using AAPM-Mayo Clinic LDCT dataset, and the trained model tested on the TCIA dataset. The proposed model outperformed all competing methods. Local noise analysis showed that noise removal was consistent across different anatomical regions, achieving 37.14% in the lung, 48.75% in soft tissue, and 35.18% in bone. Visual results also confirmed significant improvements in noise removal, preservation of structural details, and reduction of non-residual distortions. Furthermore, the model demonstrated its ability to generalize under domain shift.Significance.This work presents a framework for anatomically adapted noise removal by linking feature modification with the physical properties of noise in LDCT through a dual-modulation process for both general and tissue-related noise. The model achieves a balance between noise removal and preservation of anatomical detail, making it a robust approach to LDCT noise removal.

References

1. 1.Alfattama, S., & Vaish, A. (2026). Anatomically adaptive feature-wise linear modulation for deep learning-based low-dose CT denoising. Physics in Medicine and Biology. https://doi.org/10.1088/1361-6560/ae88af