Sensitivity of radiomic features to inter-observer variability and image pre-processing in Apparent Diffusion Coefficient (ADC) maps of cervix cancer patients

Abstract

a b s t r a c tPurpose: The aims of this study are to evaluate the stability of radiomic features from Apparent Diffusion Coefficient (ADC) maps of cervical cancer with respect to: (1) reproducibility in inter-observer delineation, and (2) image pre-processing (normalization/quantization) prior to feature extraction. Materials and methods: Two observers manually delineated the tumor on ADC maps derived from pretreatment diffusion-weighted Magnetic Resonance imaging of 81 patients with FIGO stage IB-IVA cervical cancer. First-order, shape, and texture features were extracted from the original and filtered images considering 5 different normalizations (four taken from the available literature, and one based on urine ADC) and two different quantization techniques (fixed-bin widths from 0.05 to 25, and fixed-bin count). Stability of radiomic features was assessed using intraclass correlation coefficient (ICC): poor (ICC < 0.75); good (0.75 ICC 0.89), and excellent (ICC ! 0.90). Dependencies of the features with tumor volume were assessed using Spearman's correlation coefficient (q). Results: The approach using urine-normalized values together with a smaller bin width (0.05) was the most reproducible (428/552, 78% features with ICC ! 0.75); the fixed-bin count approach was the least (215/552, 39% with ICC ! 0.75). Without normalization, using a fixed bin width of 25, 348/552 (63%) of features had an ICC ! 0.75. Overall, 26% (range 25-30%) of the features were volume-dependent (q ! 0.6). None of the volume-independent shape features were found to be reproducible. Conclusion: Applying normalization prior to features extraction increases the reproducibility of ADCbased radiomics features. When normalization is applied, a fixed-bin width approach with smaller widths is suggested. Ó 2019 The Author(s). Published by Elsevier B.V. Radiotherapy and Oncology 143 (2020) 88-94 This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).Cervical cancer is the fourth most frequent cancer in women, with an estimated 570,000 new cases in 2018, representing 6.6% of all female cancers worldwide. Cervical cancer still represents a significant burden for middle-and low-income countries [1]. Standard treatment for locally advanced (stage IB2-IVA) cervical cancer is concurrent chemoradiation.Computed Tomography (CT) and Magnetic Resonance (MR) are the standard imaging modalities for cervical cancer staging and evaluation of treatment response. Through appropriate choice of pulse sequences MR imaging provides greater soft tissue contrast than CT and enables assessment of physiological parameters and biochemical function. Diffusion-weighted imaging (DWI) in MR enables measurement of water diffusivity via generation of Apparent Diffusion Coefficient (ADC) maps, and ADC is an established biomarker of tumor cell density and related changes posttherapy [2]. DWI is increasing acquired in addition to T2weighted MRI to detect cervical tumor [3], and pre-treatment tumor ADC has been shown to be p...

Reliability of CT radiomic features reflecting tumour heterogeneity according to image quality and image processing parameters

“…As expected, this left-sided shift was not observed with the quantization range of mean ± 2 SD or mean ± 3 SD, because the effect of the outliers truncated with these settings. Although the quantization range of mean ± 3 SD was used for outlier removal 30 and provided more reliable information, the effects of outliers could not be fully removed by the normalization process, as demonstrated in the current study. Therefore, to achieve reliable RFs, outlier control should be carefully performed as part of the image segmentation process.…”

“…Outlier control is an important factor influencing RFs 29 , although to our knowledge, there have been only limited studies evaluating how outliers affect the reliability of RFs 16,30 . Our simulation results demonstrated that the unintended inclusion of outliers within ROIs seems to have the strongest impact on the GLCM and GLRLM.…”

“…The three different quantization ranges evaluated were min-max, mean ± 2 SD and mean ± 3 SD, where min-max represented the full range from the minimum to maximum values of the grey levels inside the ROIs (all grey levels were included), and the mean ± 2 SD and mean ± 3 SD options excluded grey levels outside the range of mean ± 2 (or 3) SD. The fixed bin number method in which grey level intensities are discretized to a fixed number of bins allows for direct comparisons of RFs across multiple ROIs 30,32 . Based on our literature review 10,20 , the three widely used bin numbers of 32, 64 and 128 were assigned.…”

Combination of Radiomics and Machine Learning with Diffusion-Weighted MR Imaging for Clinical Outcome Prognostication in Cervical Cancer

“…Few studies show the jarring results concerning the prognostic value of ADC in cervical cancer, some documenting poorer prognosis with a lower ADC value [59]. The VOI delineated on the T2W image could have been extrapolated on the ADC maps, to extract the radiomic features on ADC images as done in previous studies [60], but the additional normalization and binning done was beyond the scope of this study. Undeniably, a better understanding of the core spatial heterogeneity could be offered by VOI delineation and analysis of ADC imaging; however, the above factors with resource and time constrains prevented us from exploring this particular aspect, thus limiting the usefulness into routine clinical practice [61].…”

Radiomics of apparent diffusion coefficient maps to predict histologic grade in squamous cell carcinoma of the oral tongue and floor of mouth: a preliminary study

“…Therefore, the intratumor heterogeneity regarding water diffusion is deserved to be explored by radiomics based on DWI with multiple b values. A recent study reported that 63% of ADC-based radiomics features without image normalization were reported to have good inter-observer agreement in patients with cervical cancer (31). In the present study, we found a higher proportion of radiomics features (83%) with good inter-observer agreement, which may be attributable to differences in tumor type and size and image quality.…”