School of Biomedical Engineering, Southern Medical University/ Guangdong Provincial Key Laboratory of Medical Image Processing, Guangzhou 510515, China
Objective We propose a low-dose CT image restoration method based on central guidance and alternating optimization (FedGP). Methods The FedGP framework revolutionizes the traditional federated learning model by adopting a structure without a fixed central server, where each institution alternatively serves as the central server. This method uses an institution-modulated CT image restoration network as the core of client-side local training. Through a federated learning approach of central guidance and alternating optimization, the central server leverages local labeled data to guide client-side network training to enhance the generalization capability of the CT imaging model across multiple institutions. Results In the low-dose and sparse-view CT image restoration tasks, the FedGP method showed significant advantages in both visual and quantitative evaluation and achieved the highest PSNR (40.25 and 38.84), the highest SSIM (0.95 and 0.92), and the lowest RMSE (2.39 and 2.56). Ablation study of FedGP demonstrated that compared with FedGP(w/o GP) without central guidance, the FedGP method better adapted to data heterogeneity across institutions, thus ensuring robustness and generalization capability of the model in different imaging conditions. Conclusions FedGP provides a more flexible FL framework to solve the problem of CT imaging heterogeneity and well adapts to multi-institutional data characteristics to improve generalization ability of the model under diverse imaging geometric configurations.
为了验证和评估FedGP方法在多机构协作的有效性,本文使用了来自多个机构且具备不同成像几何设置的CT数据进行实验。具体来说,我们从之前的正常剂量CT图像中提取了低剂量的正弦图数据,通过仿真获得了涵盖4种成像几何配置的CT数据集。本文使用4个公开可用的数据集的胸腹部CT图像进行数据仿真,评估所提出的FedGP方法在医疗CT图像恢复任务中的性能。数据集涵盖了不同供应商和扫描仪的CT数据,包括:(1)Site #1的数据集来自于2016 NIH-AAPM-Mayo低剂量CT挑战赛数据集,包含10例患者的CT重建图像,这些患者在正常剂量(即120 kVp和200 mAs)辐射下从胸部到腹部进行扫描[24];(2) Site #2数据集来自于LDCT image and projection data,包含来自患者检查的CT投影数据,使用GE Discovery CT750 HD扫描仪[25];(3) Site #3来自于Abdomen-1K数据集,包含由飞利浦CT扫描仪采集的腹部多层kVp和有效mAs的CT图像[26];(4) Site #4来自于数据集来自CT Spine1K,包含由东芝CT扫描仪采集的不同kVp和有效mAs水平的CT图像[27]。CT仿真和重建由Astra工具箱进行[28],在实验中,我们分别使用低管电流和稀疏角度两种不同的实验设置,以验证FedGP框架在不同条件下的低剂量CT图像恢复效果。具体而言,低管电流数据模拟了不同管电流设置下的低光子量条件,而稀疏角度数据则通过减少投影角度模拟了临床中可能出现的欠采样情况。
Berrington de GonzálezA, DarbyS. Risk of cancer from diagnostic X-rays: estimates for the UK and 14 other countries[J]. Lancet, 2004, 363(9406): 345-51.
[2]
BrennerDJ, HallEJ. Computed tomography: an increasing source of radiation exposure[J]. N Engl J Med, 2007, 357(22): 2277-84.
[3]
MuraseK, NanjoT, IiS, et al. Effect of X-ray tube current on the accuracy of cerebral perfusion parameters obtained by CT perfusion studies[J]. Phys Med Biol, 2005, 50(21): 5019-29.
[4]
MengMQ, WangYB, ZhuMM, et al. Dual-task learning for low-dose CT simulation and denoising[C]//7th International Conference on Image Formation in X-Ray Computed Tomography. June 12-16, 2022. Baltimore, USA. SPIE, 2022: 602-607.
[5]
LiDY, BianZY, LiS, et al. Noise characteristics modeled unsupervised network for robust CT image reconstruction[J]. IEEE Trans Med Imaging, 2022, 41(12): 3849-61.
[6]
YouCY, LiG, ZhangY, et al. CT super-resolution GAN constrained by the identical, residual, and cycle learning ensemble (GAN-CIRCLE)[J]. IEEE Trans Med Imaging, 2020, 39(1): 188-203.
[7]
HeJ, WangYB, MaJH. Radon inversion via deep learning[J]. IEEE Trans Med Imaging, 2020, 39(6): 2076-87.
[8]
XiaWJ, LuZX, HuangYQ, et al. MAGIC: manifold and graph integrative convolutional network for low-dose CT reconstruction[J]. IEEE Trans Med Imag, 2021, 40(12): 3459-72.
[9]
HanY, YeJC. Framing U-Net via deep convolutional framelets: application to sparse-view CT[J]. IEEE Trans Med Imaging, 2018, 37(6): 1418-29.
[10]
KimB, ShimH, BaekJ. A streak artifact reduction algorithm in sparse-view CT using a self-supervised neural representation[J]. Med Phys, 2022, 49(12): 7497-515.
[11]
LeeM, KimH, KimHJ. Sparse-view CT reconstruction based on multi-level wavelet convolution neural network[J]. Phys Med, 2020, 80: 352-62.
[12]
VoigtP, von dem BusscheA. The EU general data protection regulation (GDPR): a practical guide[M]. Cham, Switzerland: Springer International Publishing, 2017.
[13]
ShellerMJ, EdwardsB, Anthony ReinaG, et al. Federated learning in medicine: facilitating multi-institutional collaborations without sharing patient data[J]. Sci Rep, 2020, 10(1): 12598.
SolanesA, PalauP, ForteaL, et al. Biased accuracy in multisite machine-learning studies due to incomplete removal of the effects of the site[J]. Psychiatry Res Neuroimaging, 2021, 314: 111313.
[20]
PerezE, StrubF, De VriesH, et al. FiLM: visual reasoning with a general conditioning layer[J]. Proc AAAI Conf Artif Intell, 2018, 32(1): 3942-51.
[21]
AlbarqouniS, BakasS, KamnitsasK, et al. Domain adaptation and representation transfer, and distributed and collaborative learning: second MICCAI Workshop, DART 2020, and first MICCAI Workshop, DCL 2020, held in conjunction with MICCAI 2020, Lima, Peru, October 4-8, 2020 , Proceedings[M]. Cham, Switzerland: Springer, 2020.
[22]
ZhuHY, XuJJ, LiuSQ, et al. Federated learning on non-IID data: a survey[J]. Neurocomputing, 2021, 465: 371-90.
[23]
LiXX, JiangMR, ZhangXF, et al. FedBN: federated learning on non-IID features via local batch normalization[EB/OL]. 2021: 2102.07623.
[24]
McColloughC. TU-FG-207A-04: overview of the low dose CT grand challenge[J]. Med Phys, 2016, 43(6Part35): 3759-60.
[25]
MoenTR, ChenBY, et al. Low-dose CT image and projection dataset[J]. Med Phys, 2021, 48(2): 902-11.
[26]
MaJ, ZhangY, GuS, et al. AbdomenCT-1K: is abdominal organ segmentation a solved problem[J]? IEEE Trans Pattern Anal Mach Intell, 2022, 44(10): 6695-714.
[27]
DengY, WangC, HuiY, et al. CTSpine 1K: a large-scale dataset for spinal vertebrae segmentation in computed tomography[EB/OL]. 2021: 2105.14711.
[28]
van AarleW, PalenstijnWJ, CantJ, et al. Fast and flexible X-ray tomography using the ASTRA toolbox[J]. Opt Express, 2016, 24(22): 25129-47.
[29]
ZengD, HuangJ, BianZY, et al. A simple low-dose X-ray CT simulation from high-dose scan[J]. IEEE Trans Nucl Sci, 2015, 62(5): 2226-33.
[30]
HintonG, Van Der MaatenL. Visualizing data using t-sne journal of machine learning research[J]. Jf Mach Learn Res, 2008, 9: 2579-605.
[31]
LiT, SahuAK, ZaheerM, et al. Federated optimization in heterogeneous networks[J]. Proceed Mach Learn Sys, 2020, 2: 429-50.