Radiomics: the process and the challenges

“This multiple testing problem has been addressed in statistics in many ways; however, the most familiar, and conservative, Bonferroni corrections severely limit the power of the test in the 50,000-test experiments [81].”

“Second-order statistics or co-occurrence matrix features can be used for texture classification [64–66] and are widely applied in medical pattern recognition tasks [67–71].”

“The level set method was initially proposed by Osher and Sethian in 1988 to track moving interfaces, and it was subsequently applied across various imaging applications in the late 1990s [38].”

“The field of high-dimensional biomarker discovery and validation has evolved rapidly over the past decade since some of the earliest microarray-based results were reported [80].”

“010 , , mineable databases from radiological images [1].”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“Currently available clinical image data which may be used for radiomics study includes the Lung Image Database Consortium, the Reference Image Database to Evaluate Response to therapy in lung cancer and others [27,28].”

“Multivariate data analysis tools such as PCA [89] and partial least squares projection to latent structures [90] (PLS) can be used to analyze quantitative features together with additional data.”

“have pioneered the work, studied various cross-validation methods and proposed unbiased error estimation called the bootstrap [93,94].”

“False discovery rates [82–84] have been developed to provide more reasonable error estimates.”

“Different feature selection methods can be used for this purpose and may exploit machine learning or statistical approaches [72–76].”

“Furthermore, gray level run length features, derived from run length matrices and using run length metrics as proposed by Galloway [66], can be extracted.”

“Many metrics have been used, like volume, center of volume and maximum surface distance, to compare characteristics like robustness and accuracy [57,58].”

“DWI has been proposed as a cancer imaging biomarker, and there are efforts to develop quality control protocols [18,19].”

“False discovery rates [82–84] have been developed to provide more reasonable error estimates.”

“Before a scanner is used in a quantitative manner, interinstitution cross-calibration and quality control such as proposed recently [14] are necessary (Fig.”

“The Jaccard Similarity Index (SI) is the measure of the overlap of two or more volumes and is calculated as the ratio of voxel-wise intersection to union of target and reference images [59]:”

“Second-order statistics or co-occurrence matrix features can be used for texture classification [64–66] and are widely applied in medical pattern recognition tasks [67–71].”

“Following the best practices in developing and then independently validating the observations in a distinct cohort is essential for reproducible results [92].”

“This cluster heat map is a synthesis of various graphic displays developed by statisticians over more than a century [86].”

“PET–CT Quantitative imaging with 2-deoxy-2-[(18)F]fluoro-D-glucose (18-FDG) PET scans is a challenge because it not only requires calibration of the scanner and standardization of the scan protocol but also requires the patient and staff to adhere to a strict patient protocol [14,15].”

“[91], many genomics-based studies that have been published contain significant analytical errors.”

“Second-order statistics or co-occurrence matrix features can be used for texture classification [64–66] and are widely applied in medical pattern recognition tasks [67–71].”

“Following a complete manual extraction of numerous (N100) image features, a subset of 14 features was able to predict 80% of the gene expression pattern in hepatocellular carcinoma using computed tomographic (CT) images [12].”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“, mean, median, min, max, range, skewness, kurtosis), but also more complex values, such as metabolic volume above an absolute SUV of 5 or the fraction of highdensity tissue measured with CT [60,61].”

“A similar extraction of features from contrast-enhanced magnetic resonance images (MRI) of glioblastoma was able to predict immunohistochemically identified protein expression patterns [13].”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“Dimensionality reduction can also be achieved by combining or transforming the original features to obtain a new set of features by using methods like principal component analysis (PCA) [73].”

“Approaches we have used include neural networks [87], linear regression [87] and Cox proportional hazards regression [88].”

“Results of the DCE MRI depend on the contrast agent dose, method of administration, pulse sequence used, field strength of the scanner and the analysis method used [20–22].”

“[RECIST]) or two-dimensional (2D) (World Health Organization) long-axis measures [2].”

“In focused studies, texture features have been shown to provide significantly higher prognostic power than ROI-based methods [8–11].”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“Inference in small samples has seen renewed interest with the advent of genomics technologies, especially in classification [95].”

“There are efforts to develop a standardized lexicon for the description of such lesions [4,5] and to include these descriptors via annotated image markup into quantitative, mineable data [6,7].”

“Therefore, a new data warehouse model that incorporates the metadata layer into the data model, including a comprehensive data dictionary along with calculated data quality attributes such as completeness, accuracy and consistency, has been utilized for the radiomics project [79].”

“Many metrics have been used, like volume, center of volume and maximum surface distance, to compare characteristics like robustness and accuracy [57,58].”

“It has been observed that enhancement in individual tumors can be heterogeneous and that analysis of this heterogeneity has prognostic value [9].”

“Different feature selection methods can be used for this purpose and may exploit machine learning or statistical approaches [72–76].”

“Results of the DCE MRI depend on the contrast agent dose, method of administration, pulse sequence used, field strength of the scanner and the analysis method used [20–22].”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“In the majority of cases, manual segmentation tends to overestimate the lesion volume to ensure the entire lesion is identified [54], and the process is highly variable [55,56].”

“Manually traced segmentations are often used as gold standard or ground truth [30] against which the accuracy of the automatic segmentation is evaluated.”

“One of the largest online CT image repositories is the National Biomedical Image Archive (NBIA) hosted by the National Cancer Institute.”

“Different investigators use different methods to convert DCE MRI signal intensities to contrast agent concentration [23–25].”

“Currently available clinical image data which may be used for radiomics study includes the Lung Image Database Consortium, the Reference Image Database to Evaluate Response to therapy in lung cancer and others [27,28].”

“In addition, most popular error estimates has been studied in context of small sample classification [97].”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“There has been extensive studies to make unbiased inference in small samples, one approach was to create synthetic samples following the distribution of the sample groups and report errors of the newly formed population [96].”

“Quantitative features describing the geometric shape of a tumor can also be extracted from the 3D surface of the rendered volumes [62].”

“In the majority of cases, manual segmentation tends to overestimate the lesion volume to ensure the entire lesion is identified [54], and the process is highly variable [55,56].”

“From a technical viewpoint, the main challenges are the dose calibration and the metabolic volume or volume-of-interest (VOI) reconstruction that depends heavily on the scan protocol and source-to-background ratio [16].”

“DWI has been proposed as a cancer imaging biomarker, and there are efforts to develop quality control protocols [18,19].”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“There are efforts to develop a standardized lexicon for the description of such lesions [4,5] and to include these descriptors via annotated image markup into quantitative, mineable data [6,7].”

“These selected features can also be used to develop classifier models based on machine learning algorithms to improve the performance [29].”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“Different investigators use different methods to convert DCE MRI signal intensities to contrast agent concentration [23–25].”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“, mean, median, min, max, range, skewness, kurtosis), but also more complex values, such as metabolic volume above an absolute SUV of 5 or the fraction of highdensity tissue measured with CT [60,61].”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“Different investigators use different methods to convert DCE MRI signal intensities to contrast agent concentration [23–25].”

“Results of the DCE MRI depend on the contrast agent dose, method of administration, pulse sequence used, field strength of the scanner and the analysis method used [20–22].”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“There are efforts to develop a standardized lexicon for the description of such lesions [4,5] and to include these descriptors via annotated image markup into quantitative, mineable data [6,7].”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“From a patient protocol perspective, administration issues (residual activity in syringe, paravenous administration), blood glucose level [17], uptake period, breathing, patient comfort and inflammation all influence the quantitation of the standardized uptake value (SUV) of 18-FDG.”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“These measures do not reflect the complexity of tumor morphology or behavior, nor, in many cases, are changes in these measures predictive of therapeutic benefit [3].”

“Many popular segmentation algorithms have been applied in medical imaging studies within the last 20 years; the most popular ones include region-growing methods [31,32], level set methods [33–38], graph cut methods [39–44], active contours (snake) algorithms [45–49] and semiautomatic segmentations such as livewires [5…”

“In the majority of cases, manual segmentation tends to overestimate the lesion volume to ensure the entire lesion is identified [54], and the process is highly variable [55,56].”

“Second-order statistics or co-occurrence matrix features can be used for texture classification [64–66] and are widely applied in medical pattern recognition tasks [67–71].”

“5 iv Chapter 1 On this document While analysis of medical images has practically taken place since the first image was recorded, high throughput analysis of medical images is a more recent phenomenon (Kumar et al., 2012; Lambin et al., 2012; Gillies et al., 2015). The aim of such a radiomics process is to provide decis…”