ComBat harmonization for radiomic features in independent phantom and lung cancer patient computed tomography datasets

PURPOSE To evaluate the Combatting Batch effect (ComBat) harmonization algorithm's ability to reduce the variation in radiomic features arising from different imaging protocols and independently verify published results. METHODS The Gammex computed tomography (CT) electron density phantom and Quasar body phantom were imaged using 32 different chest imaging protocols. 107 radiomic features were extracted from 15 spatially varying spherical contours between 1.5cm and 3cm in each of the lung300 density, lung450 density, and wood inserts. The Kolmogorov-Smirnov test was used to determine significant differences in the distribution of the features and the Concordance Correlation Coefficient (CCC) was uses to measure the repeatability of the features from each protocol variation class (kVp, pitch, etc.) before and after ComBat harmonization. P-values were corrected for multiple comparisons using the Benjamini-Hochberg-Yekutieli procedure. Finally, the ComBat algorithm was applied to human subject data using 6 different thorax imaging protocols with 135 patients. Spherical contours of un-irradiated lung (2cm) and vertebral bone (1cm) were used for radiomic feature extraction. RESULTS ComBat harmonization reduced the percentage of features from significantly different distributions to 0-2% or preserved 0% across all protocol variations for the lung300, lung450 and wood inserts. For the human subject data, ComBat harmonization reduced the percentage of significantly different features from 0-59% for bone and 0-19% for lung to 0% for both. CONCLUSION This work verifies previously published results and demonstrates that ComBat harmonization is an effective means to harmonize radiomic features extracted from different imaging protocols to allow comparisons in large multi-institution datasets. Biological variation can be explicitly preserved by providing the ComBat algorithm with clinical or biological variables to protect. ComBat harmonization should be tested for its effect on predictive models.

[1]  Peter Balter,et al.  Delta-radiomics features for the prediction of patient outcomes in non–small cell lung cancer , 2017, Scientific Reports.

[2]  Y. Benjamini,et al.  THE CONTROL OF THE FALSE DISCOVERY RATE IN MULTIPLE TESTING UNDER DEPENDENCY , 2001 .

[3]  Ragini Verma,et al.  Harmonization of multi-site diffusion tensor imaging data , 2017, NeuroImage.

[4]  Andriy Fedorov,et al.  Computational Radiomics System to Decode the Radiographic Phenotype. , 2017, Cancer research.

[5]  Henning Müller,et al.  Learning cross-protocol radiomics and deep feature standardization from CT images of texture phantoms , 2019, Medical Imaging.

[6]  M. Cecchini,et al.  Ultrastructural Characterization of the Lower Motor System in a Mouse Model of Krabbe Disease , 2016, Scientific Reports.

[7]  Mohan M. Trivedi,et al.  Segmentation of a high-resolution urban scene using texture operators , 1984, Comput. Vis. Graph. Image Process..

[8]  Geoffrey G. Zhang,et al.  Intrinsic dependencies of CT radiomic features on voxel size and number of gray levels , 2017, Medical physics.

[9]  Sang Joon Park,et al.  Impact of Reconstruction Algorithms on CT Radiomic Features of Pulmonary Tumors: Analysis of Intra- and Inter-Reader Variability and Inter-Reconstruction Algorithm Variability , 2016, PloS one.

[10]  M. Soussan,et al.  A Postreconstruction Harmonization Method for Multicenter Radiomic Studies in PET , 2018, The Journal of Nuclear Medicine.

[11]  P. Lambin,et al.  Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach , 2014, Nature Communications.

[12]  W. Tsai,et al.  Reproducibility of radiomics for deciphering tumor phenotype with imaging , 2016, Scientific Reports.

[13]  Michael A Jacobs,et al.  Integrated radiomic framework for breast cancer and tumor biology using advanced machine learning and multiparametric MRI , 2017, npj Breast Cancer.

[14]  Cheng Li,et al.  Adjusting batch effects in microarray expression data using empirical Bayes methods. , 2007, Biostatistics.

[15]  Geoffrey G. Zhang,et al.  Voxel size and gray level normalization of CT radiomic features in lung cancer , 2018, Scientific Reports.

[16]  Jörg Hausleiter,et al.  Quantitative Evaluation of the Performance of a New Test Bolus–Based Computed Tomographic Angiography Contrast-Enhancement–Prediction Algorithm , 2015, Investigative radiology.

[17]  Fanny Orlhac,et al.  Validation of A Method to Compensate Multicenter Effects Affecting CT Radiomics. , 2019, Radiology.

[18]  Laurence Court,et al.  Harmonizing the pixel size in retrospective computed tomography radiomics studies , 2017, PloS one.

[19]  Lawrence H. Schwartz,et al.  Assessing Agreement between Radiomic Features Computed for Multiple CT Imaging Settings , 2016, PloS one.

[20]  L. Lin,et al.  A concordance correlation coefficient to evaluate reproducibility. , 1989, Biometrics.

[21]  Russell T. Shinohara,et al.  Harmonization of cortical thickness measurements across scanners and sites , 2017, NeuroImage.

[22]  Jinzhong Yang,et al.  NSCLC tumor shrinkage prediction using quantitative image features , 2016, Comput. Medical Imaging Graph..

[23]  P. F. Vasconcelos,et al.  In situ immune response and mechanisms of cell damage in central nervous system of fatal cases microcephaly by Zika virus , 2018, Scientific Reports.

[24]  H. Aerts,et al.  Applications and limitations of radiomics , 2016, Physics in medicine and biology.