Vol.:(0123456789)

Radiological Physics and Technology (2018) 11:265–273 https://doi.org/10.1007/s12194-018-0462-5

Computer-aided diagnosis with radiogenomics: analysis of the relationship between genotype and morphological changes of the brain magnetic resonance images

Chiharu Kai1 · Yoshikazu Uchiyama2  · Junji Shiraishi2 · Hiroshi Fujita3 · Kunio Doi4,5

Received: 2 October 2017 / Revised: 3 May 2018 / Accepted: 4 May 2018 / Published online: 10 May 2018 © Japanese Society of Radiological Technology and Japan Society of Medical Physics 2018

AbstractIn the post-genome era, a novel research field, ‘radiomics’ has been developed to offer a new viewpoint for the use of genotypes in radiology and medicine research which have traditionally focused on the analysis of imaging phenotypes. The present study analyzed brain morphological changes related to the individual’s genotype. Our data consisted of magnetic resonance (MR) images of patients with mild cognitive impairment (MCI) and Alzheimer’s disease (AD), as well as their apolipoprotein E (APOE) genotypes. First, statistical parametric mapping (SPM) 12 was used for three-dimensional ana-tomical standardization of the brain MR images. A total of 30 normal images were used to create a standard normal brain image. Z-score maps were generated to identify the differences between an abnormal image and the standard normal brain. Our experimental results revealed that cerebral atrophies, depending on genotypes, can occur in different locations and that morphological changes may differ between MCI and AD. Using a classifier to characterize cerebral atrophies related to an individual’s genotype, we developed a computer-aided diagnosis (CAD) scheme to identify the disease. For the early detection of cerebral diseases, a screening system using MR images, called Brain Check-up, is widely performed in Japan. Therefore, our proposed CAD scheme would be used in Brain Check-up.

Keywords Mild cognitive impairment · Alzheimer’s disease · Radiomics · Radiogenomics · Computer-aided diagnosis

1 Introduction

As we enter the post-genome era, the technologies for genetic analysis have advanced dramatically along with an astonishing decline in the cost of analysis. In 2007, the

cost to sequence the entire human genome was 10 million USD, but decreased sharply to only 1000 USD in 2013 [1]. Therefore, it is likely that genetic testing may be performed in daily clinical practice in the near future. In this context, a novel research area, ‘radiomics’ [2–7] is being developed to offer a new viewpoint of genotypes to research in radiology and medicine which have traditionally focused on the analy-sis of imaging phenotypes. Figure 1 shows the relationship between radiomics and omics research. Gene and protein data are referred to as genome and proteome, respectively. Their analytical techniques and data obtained using this approach is called ‘genomics’ and ‘proteomics’ with—omics attached at the end. Hence, radiomics is a generic term to refer to the analytical technologies of medical images and the image features obtained by this approach. In particular, research on the relationship between medical images and the genome is called ‘radiogenomics’; similarly, ‘radioproteom-ics’ refers to research on the relationship between medical images and proteins.

With the arrival of a worldwide aging society, the inci-dence of dementia tends to increase. Therefore, this study

* Yoshikazu Uchiyama y_uchi@kumamoto-u.ac.jp

1 Graduate School of Health Sciences, Kumamoto University, 4-24-1 Kuhonji, Chuo-ku, Kumamoto-shi, Kumamoto 862-0976, Japan

2 Department of Medical Physics, Faculty of Life Sciences, Kumamoto University, 4-24-1 Kuhonji, Chuo-ku, Kumamoto-shi, Kumamoto 862-0976, Japan

3 Department of Intelligent Image Information, Graduate School of Medicine, Gifu University, 1-1 Yanagido, Gifu-shi, Gifu 501-1194, Japan

4 Department of Radiology, The University of Chicago, 5841 South Maryland Avenue, Chicago, IL 60637, USA

5 Gunma Prefectural College of Health Sciences, 323-1 Kamiokimachi, Maebashi-shi, Gunma 371-0052, Japan

266 C. Kai et al.

focused on computer-aided diagnosis (CAD) [8–12] for the early detection of Alzheimer’s disease (AD). Several methods have been proposed for the early detection of AD, including the automatic segmentation of white and gray brain matter in MR images [13], extraction of cerebrospinal fluid [14], and measurement of the thickness of the brain cortex [15]. In addition, a voxel-based specific regional analysis system for Alzheimer’s disease (VSRAD) [16], based on voxel-based morphometry (VBM) is used in clini-cal settings for the evaluation of brain atrophy. However, these methods are used to quantify image features, with a focus on morphological changes related to cerebral atrophy, but their relationship to genes have not yet been analyzed.

In a related study using image features and genes, Muñoz-Ruiz et al. [17, 18] proposed a computer-assisted clinical decision-making tool for the differential diagnosis of fron-totemporal dementia and AD. The study used available diag-nostic biomarkers from neuropsychological test, volumetric magnetic resonance imaging, single-photon emission tomog-raphy, cerebrospinal fluid biomarkers, and the APOE geno-type. The use of image features and genotype is the same as that in the present study. However, this related study would be classified as CAD for differential diagnosis.

In this study, we investigated the relationship between APOE genotypes and morphological changes in cerebral atrophy. APOE is associated with AD and an individual’s APOE genotype can be determined by blood tests. If the

development of a lesion is related to an individual genotype, it may be possible to develop a CAD scheme for the early detection of the disease by identifying image features related to the process by which a lesion is generated. This paper is organized as follows: we propose a method for the quantita-tive evaluation of brain morphological changes in Sect. 3; we clarify the differences in the morphological changes of brain atrophy related to genotypes and develop a CAD using genotype and image features in Sect. 4; and we discuss dif-ferences between the present study and radiomic research in Sect. 5.

2 Materials

In this study, MR images and APOE genotypes of 206 cases were selected from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database [19]. The images were acquired with a 3.0T MR device by magnetization-prepared rapid acquisition with gradient echo (MP-RAGE). The matrix size was 256 × 256 with a voxel size of 1.0 × 1.0 × 1.2 mm3. This data included 56 normal subjects as well as 81 and 69 patients with mild cognitive impairment (MCI) and AD, respectively. Figure 2 shows a model of the clinical trajec-tory of AD [20]. MCI is considered a preliminary stage of AD, with 10–15% of patients with MCI reportedly develop-ing AD every year [21]. Therefore, we selected MR images

Fig. 1 Relationship between radiomics and omics research. The images are cited from reference [2]. Medical images may record multi-level cellular activities from the molecular level to the forma-

tion of a lesion. Therefore, medical images are inclusive in the rela-tionship with omics information of each hierarchy

267Computer-aided diagnosis with radiogenomics: analysis of the relationship between genotype…

of patients with MCI and AD, to quantify image features for the transition of the disease state from MCI to AD. In addi-tion, we used APOE gene in this study because it is known to be related to AD [22, 23]. The APOE genotypes include 2/2, 2/3, 2/4, 3/3, 3/4, and 4/4. Since few people have the APOE ε2 (2/2) and those with the APOE ε4 genotype tend to develop AD, we included those patients with the APOE ε3 (3/3) and APOE ε4 (4/4) genotypes in this study too.

3 Methods

Figure 3 shows the flowchart of our method. To analyze the morphological changes in the brain, the brain images were normalized for size and anatomical position. Average image and standard deviation images were prepared using 30 normal cases. The abnormal cases were divided into four groups: MCI with APOE ε3, AD with APOE ε3, MCI with APOE ε4, and AD with APOE ε4. Four average images were created from each of these four groups. We created Z-score maps by normalizing the average images of the four groups to the normal average and the normal standard deviation (SD) images. The degree of atrophy of each abnormal case was quantitatively analyzed using the pixel values of the Z-score map. The details of our method are described below.

3.1 Standardization of brain anatomy

We implemented standardization of the brain anatomy using statistical parametric mapping (SPM) 12 [24] as a preproc-essing step for the analysis of morphological changes in the brain. First, realignment processing was used to correct the head movement due to body motion and pulsation. A normali-zation procedure was subsequently used to convert the image to match the standard brain of the Montreal Neurological Insti-tute (MNI) prepared in SPM. Additionally, a linear grayscale transformation was used to correct the image density. Based on the image histograms, a pixel value of 0.05% at the top was

obtained by applying the P-tile method. The linear grayscale transformation was made so that the pixel values equal to or greater than the value obtained by P-tile method were set as 255, and the minimum value was 0. These processes were used for standardization of the brain anatomy; the anatomical posi-tion was the same for all images. We could, therefore, analyze the morphological changes by comparing pixel values at the corresponding positions.

3.2 Z‑score maps

We created Z-score maps to visually analyze changes in brain morphology in the four groups classified by MCI and AD sta-tus and APOE genotypes. First, a normal average image and a normal SD image were prepared using 30 standardized MR images from unaffected patients. These images were obtained by calculating the average value and SD in each voxel. Next, the average MR images were also prepared for each of the four abnormal groups. The following equation was used to create the four Z-score maps:

(1)ZScore (x, y, z) =Input (x, y, z) −Mean (x, y, z)

SD (x, y, z),

Fig. 2 Model of the clinical trajectory of AD, which is modified form Fig. 1 in reference [18]. Starting treatment during MCI may suppress the progression to AD

Fig. 3 Overall scheme for determining the Z-score map of each patient according to their genotype

268 C. Kai et al.

where Mean (x, y, z) and SD (x, y, z) are the normal average image and normal SD image, respectively. Input (x, y, z) is the average image for each abnormal group. Because the voxel value of Z-score map can show the normalized differ-ence from the normal brain, cerebral atrophy would result in a negative Z-score value.

3.3 Quantitative evaluation of the degree of cerebral atrophy

By setting the region of interest (ROI) in the normal standard brain, the degree of atrophy in the ROI can be quantified. In this study, three-dimensional ROIs were set in three regions; i.e., the hippocampus, occipital lobe, and islet cortex. We developed a software for filling ROIs in three dimensions using C# language, and it was used for marking the ROIs. We determined the total value in the ROI when the Z-score was negative, as the degree of cerebral atrophy in the region. Box-whisker charts were produced to evaluate the differ-ences in the degree of atrophy among the four abnormal and the normal groups.

3.4 Construction of CAD using genotype and image features

To demonstrate the possibility of constructing a CAD using genotype and image features, we need to add a classifier. However, the number of samples decreased (especially in normal cases), when we divided all cases by genotype. Hence, we could not use complicated classifiers such as arti-ficial neural network, support vector machine, and so on. In this study, therefore, we employed quadratic discriminant analysis (QDA) as a classifier for distinguishing between normal and abnormal cases. Since QDA uses the averages and variances of normal and abnormal cases for generat-ing the decision boundary of a quadratic surface, training and testing can be performed with a small number of sam-ples. The input data of QDA was a genotype (APOEε3 or APOEε4) and the three image features (the degree of atrophy of three regions, i.e., the hippocampus, occipital lobe, and islet cortex). The output classes of QDA were normal, MCI, and AD, respectively. For evaluating the performance of CAD, we employed a cross validation test using the ‘leave-one-patient-out’ method.

4 Results

Figure 4 shows the Z-score maps of the average images obtained for each of the four abnormal groups; i.e., MCI with APOE ε3, AD with APOE ε3, MCI with APOE ε4,

and AD with APOE ε4. The gray scale in the Z-score maps indicates pixel values in the range for the normal standard brain. The green represents the anatomical locations of cerebral atrophy where the pixel value was lower than that of the normal brain. As shown in Fig. 4, cerebral atrophy was observed in the parietal lobes of MCI patients with the APOE ε4 genotype, but not in those with the APOE ε3 genotype. In contrast, the cerebral atrophy in AD patients with the APOE ε3 genotype was spread throughout the brain. Cerebral atrophy tended to remain localized in AD patients with APOE ε4. In terms of genotype, sharp cer-ebral atrophy was observed from MCI to AD in patients with APOE ε3. However, rapid brain changes were not seen from MCI to AD in patients with APOE ε4; the cerebral atrophy remained in the hippocampal region of patients with AD. These results demonstrate that the site of cerebral atrophy and the formation process of the dis-ease were different depending on the genotype.

To investigate the individual variation, the hippocam-pus, occipital lobe, and islet cortex regions were marked in three dimensions and the degree of cerebral atrophy in each region was determined. Figure 5 shows box-whisker charts indicating the degree of cerebral atrophy in each region. Twenty-six normal cases, which were not used to create the normal standard brain, were used to con-struct the box-whisker plots of normal cases. As shown in Fig. 5, individual variation was large in the hippocampus. However, there were differences in the degree of cerebral atrophy depending on the patient genotypes in all three regions.

To demonstrate the possibility of constructing CAD using a combination of genotype and image features, we employed QDA for distinguishing between normal cases and abnormal cases. Table 1 shows the results of QDA using ‘leave-one-patient-out’ method. The number of normal cases is 26; the numbers of MCI and AD are 81 and 69, respectively. Hence, the total number of abnor-mal cases for both MCI and AD is 150. The sensitivity, specificity, and diagnostic accuracy for the detection of abnormal cases were 74.7% (112/150), 73.1% (19/26), and 74.4% (131/176), respectively, when we used genotype and three image features; whereas, they were 78.0% (117/150), 15.4% (4/26), and 68.8% (121/176) when we used only the three image features without genotype. Figure 6 shows the receiver operating characteristic (ROC) curves for the distinction between normal and the other cases using QDA output values for normal class. The area under the curve (AUC) with genotype and image features was 0.797, whereas with three image features it was 0.528. Although there was an inverse hook on the lower left corner due to the small sample size of normal cases, we could obtain higher AUC value ( p < 0.01 ) using a combination of geno-type and image features.

269Computer-aided diagnosis with radiogenomics: analysis of the relationship between genotype…

Fig. 4 Relationship between genotypes and cerebral atrophies in patients with MCI and AD. The upper row shows the patients with APOE ε3, and the lower row shows the patients with APOE ε4. The left shows the patients with MCI, and the right shows the patients with AD

Fig. 5 Box-whisker charts indicating the degree of cerebral atrophy in each of the three brain regions

270 C. Kai et al.

5 Discussion

Our results revealed that depending on the genotype, (1) the anatomical locations of the cerebral atrophy differ and (2) the transition in the disease state from MCI to AD differs. Mori et al. [25] reported that APOE ε4 was sig-nificantly correlated with the rate of hippocampal atrophy; this result is different from our result. On the other hand, Barber et al. [26] reported that APOE ε4 allele does not determine medial temporal lobe atrophy. Yasuda et al. [27] reported that patients with AD who carried the APOE ε4 allele had less whole brain atrophy than those without it. These results show the same tendency as our result. In addition, Mori et al. [25] mentioned above, concluded that APOE genotype is not only related to the development of AD but also involved in the pattern of disease progres-sion; therefore, the conclusion of their study is the same as our result.

Since a person’s genotype can be determined by genetic screening, we can use this information to identify patients susceptible to AD. Those patients may be subjected to

periodic image examinations for the early detection of the onset of disease. If it is possible to identify changes in cerebral atrophy related to the patient’s genotype based on image examination, the onset may be detected earlier than that at present. Therefore, if a computer can accu-rately learn image features extracted from the change in lesions related to a patient’s genotype, it would be possible to develop a CAD scheme for the early detection of dis-ease. People with the APOE ε4 genotype tend to develop AD [22]. However, the cerebral atrophy in AD patients with the APOE ε4 genotype did not spread throughout the brain and tended to remain localized, as shown in Fig. 4. Therefore, it is important to make image interpretations by taking into account the patient’s genotype. In fact, the result of Fig. 6 indicated that diagnostic accuracy was improved using genotype and image feature. The result of Table 1 would not be sufficient for clinical use. How-ever, it indicates the possibility of realization of CAD with radiogenomics.

Research on radiomics is also investigating the relation-ship between genes and image features [28–37]. For exam-ple, Huynh et al. [28] reported the relationship between radiomic features and disease recurrence in early-stage non-small cell lung cancer. Li et al. [34] assessed the rela-tionship between radiomic features extracted from MR images and the recurrence risk score output by the multigene assays commonly used for the examination of gene expres-sion in breast cancer. Yu et al. [37] proposed a method to estimate isocitrate dehydrogenase 1 (IDH1) status, which was reportedly correlated with the development, treatment, and prognosis of glioma, using radiomic features obtained from T2-weighted and fluid attenuation inversion recovery (FLAIR) images. Figure 7 shows the differences between these radiomic studies and the present study. Medical care is performed in the order of detection of lesion, differential diagnosis, and treatment. Radiomic research may be used to analyze the relationship between cancer genotype and image features and may also predict the possibility of recurrence and metastasis and suggest an optimal treatment.

Although we determined image features in three cer-ebral regions for the quantitative evaluation of the degree

Table 1 The results of QDA when we use (a) genotype and image features and (b) image features

Fig. 6 The ROC curves for the distinction between normal and the other cases (MCI and AD)

271Computer-aided diagnosis with radiogenomics: analysis of the relationship between genotype…

of cerebral atrophy, we could not obtain image features that clearly distinguished between normal and abnormal cases. To investigate the degree of cerebral atrophy in other regions, the problem of multiple test [38] must be solved; however, the sample size becomes small when normal and abnormal cases are divided by genotypes. Since a large num-ber of cases would be necessary for solving the problem of a multiple test, it can be said that this is the limitation of the present study.

6 Conclusion

The results of this study revealed that the morphological changes in the cerebral atrophy may differ according to the patient genotype. Using a classifier to learn changes in image features of the lesion related to the difference in genotype, we developed a CAD scheme to support per-sonalized medicine. For the early detection of cerebral

diseases, a screening system using MR images, called Brain Check-up, is widely performed in Japan. There-fore, our proposed CAD scheme would be used in Brain Check-up.

Acknowledgements This study was partly supported by a Grant-in-Aid for Scientific Research (C) (No. 17K09067), by the Japan Society for the Promotion of Science and a Grant-in-Aid for Scientific Research on Innovative Areas (Multidisciplinary Computational Anatomy No. 26108001) from the Ministry of Education, Culture, Sports, Science, and Technology, Japan.

Compliance with ethical standards

Conflict of interest The authors declare no conflict of interest.

Human rights All study procedures involving human participants were performed in accordance with the ethical standards of the Institutional Review Board (IRB) and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards.

Informed consent We used data from a public database in this study. The IRB of our institute allowed us to use data from those cases.

Fig. 7 Personalized medicine by the fusion of gene and image examinations. Our study proposes a different concept from that of radiomic research

272 C. Kai et al.

Animal rights This study did not involve any animal models.

References

1. Hayden EC. Technology: the $1,000 genome. Nature. 2014;507(7492):294–5.

2. Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, van Stiphout RG, Granton P, Zegers CM, Gillies R, Boellard R, Dek-ker A, Aerts HJ. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48(4):441–6.

3. Mazurowski MA. Radiogenomics: what it is and why it is impor-tant. J Am Coll Radiol. 2015;12(8):862–6.

4. Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, Bussink J, Monshouwer R, Haibe-Kains B, Rietveld D, Hoebers F, Rietbergen MM, Leemans CR, Dekker A, Quack-enbush J, Gillies RJ, Lambin P. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun. 2014;5:4006.

5. Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology. 2016;278(2):563–77.

6. Parekh V, Jacobs MA. Radiomics: a new application from established techniques. Expert Rev Precis Med Drug Dev. 2016;1(2):207–26.

7. Lee G, Lee HY, Park H, Schiebler ML, van Beek EJR, Ohno Y, Seo JB, Leung A. Radiomics and its emerging role in lung cancer research, imaging biomarkers and clinical management: state of the art. Eur J Cancer. 2017;86:297–307.

8. Doi K. Computer-aided diagnosis in medical imaging: historical review, current status and future potential. Comput Med Imaging Graph. 2007;31(4–5):198–211.

9. Doi K. Overview on research and development of com-puter-aided diagnostic schemes. Semin Ultrasound CT MR. 2004;25(5):404–10.

10. Giger ML. Computerized analysis of images in the detection and diagnosis of breast cancer. Semin Ultrasound CT MR. 2004;25(5):411–8.

11. Giger ML, Karssemeijer N, Schnabel JA. Breast image analysis for risk assessment, detection, diagnosis, and treatment of cancer. Annu Rev Biomed Eng. 2013;15:327–57.

12. Li Q, Nishikawa RM. Computer-aided detection and diagnosis in medical imaging. Florida: CRC Press; 2015.

13. Goldenberg R, Kimmel R, Rivlin E, Rudzsky M. Cortex segmen-tation: a fast variational geometric approach. IEEE Trans Med Imag. 2002;21(12):1544–51.

14. Arimura H, Yoshiura T, Kumazawa S, Tanaka K, Koga H, Mihara F, Honda H, Sakai S, Toyofuku F, Higashida Y. Auto-mated method for identification of patients with Alzheimer’s diseases based on three-dimensional MR images. Acad Radiol. 2008;15(3):274–84.

15. Jones SE, Buchbinder BR, Aharon I. Three-dimensional mapping of cortical thickness using Laplace’s equation. Hum Brain Mapp. 2000;11(1):12–32.

16. Hirata Y, Matsuda H, Nemoto K, Ohnishi T, Hirao K, Yamashita F, Asada T, Iwabuchi S, Samejima H. Voxel-based morphometry to discriminate early Alzheimer’s disease from controls. Neurosci Lett. 2005;382(3):269–74.

17. Muñoz-Ruiz M, Hartikainen P, Hall A, Mattila J, Koikkalainen J, Herukka SK, Julkunen V, Vanninen R, Liu Y, Lötjönen J, Soininen H. Disease state fingerprint in frontotemporal degen-eration with reference to Alzheimer’s disease and mild cognitive impairment. J Alzheimers Dis. 2013;35(4):727–39.

18. Muñoz-Ruiz M, Hall A, Mattila J, Koikkalainen J, Herukka SK, Husso M, Hänninen T, Vanninen R, Liu Y, Hallikainen M, Lötjönen J, Remes AM, Alafuzoff I, Soininen H, Hartikainen P. Using the disease state fingerprint tool for differential diagnosis of frontotemporal dementia and Alzheimer’s disease. Dement Geriatr Cogn Dis Extra. 2016;6(2):313–29.

19. http://adni.loni.usc.edu/. 20. Sperling RA, Aisen PS, Beckett LA, Bennett DA, Craft S, Fagan

AM, Iwatsubo T, Jack CR Jr, Kaye J, Montine TJ, Park DC, Rei-man EM, Rowe CC, Siemers E, Stern Y, Yaffe K, Carrillo MC, Thies B, Morrison-Bogorad M, Wagster MV, Phelps CH. Toward defining the preclinical stages of Alzheimer’s disease: recommen-dations from the National Institute on Aging-Alzheimer’s Asso-ciation workgroups on diagnostic guidelines for Alzheimer’s dis-ease. Alzheimers Dement. 2011;7(3):280–92.

21. Petersen RC, Smith GE, Waring SC, Ivnik RJ, Tangalos EG, Kok-men E. Mild cognitive impairment clinical characterization and outcome. Arch Neurol. 1999;56(3):303–8.

22. Corder EH, Saunders AM, Strittmatter WJ, Schmechel DE, Gaskell PC, Small GW, Roses AD, Haines JL, Pericak-Vance MA. Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer’s disease in late onset families. Science. 1993;261(5123):921–3.

23. Poirier J, Davignon J, Bouthillier D, Kogan S, Bertrand P, Gauth-ier S. Apolipoprotein E polymorphism and Alzheimer’s disease. Lancet. 1993;342(8873):697–9.

24. http://www.fil.ion.ucl.ac.uk/spm/. 25. Mori E, Lee K, Yasuda M, Hashimoto M, Kazui H, Hirono N, Mat-

sui M. Accelerated hippocampal atrophy in Alzheimer’s disease with apolipoprotein E ε4 allele. Ann Neurol. 2002;51(2):209–14.

26. Barber R, Gholkar A, Scheltens P, Ballard C, McKeith IG, Morris CM, O’Brien JT. Apolipoprotein E ε4 allele, temporal lobe atro-phy, and white matter lesions in late-life dementias. Arch Neurol. 1999;56(8):961–5.

27. Yasuda M, Mori E, Kitagaki H, Yamashita H, Hirono N, Shimada K, Maeda K, Tanaka C. Apolipoprotein E ε4 allele and whole brain atrophy in late-onset Alzheimer’s diseases. Am J Psychiatry. 1998;155(6):779–84.

28. Huynh E, Coroller TP, Narayan V, Agrawal V, Romano J, Franco I, Parmar C, Hou Y, Mak RH, Aerts HJ. Associations of radiomic data extracted from static and respiratory-gated CT scans with disease recurrence in lung cancer patients treated with SBRT. PLoS One. 2017;12(1):e0169172.

29. Huang Y, Liu Z, He L, Chen X, Pan D, Ma Z, Liang C, Tian J, Liang C. Radiomics signature: a potential biomarker for the pre-diction of disease-free survival in early-stage (I or II) non-small cell lung cancer. Radiology. 2016;281(3):947–57.

30. Wu W, Parmar C, Grossmann P, Quackenbush J, Lambin P, Bus-sink J, Mak R, Aerts HJ. Exploratory study to identify radiomics classifiers for lung cancer histology. Front Oncol. 2016;6:71.

31. Parmar C, Leijenaar RT, Grossmann P, Rios Velazquez E, Bussink J, Rietveld D, Rietbergen MM, Haibe-Kains B, Lambin P, Aerts HJ. Radiomic feature clusters and prognostic signatures specific for Lung and Head & Neck cancer. Sci Rep. 2015;5:11044.

32. Vallières M, Freeman CR, Skamene SR, El Naqa I. A radiomics model from joint FDG-PET and MRI texture features for the pre-diction of lung metastases in soft-tissue sarcomas of the extremi-ties. Phys Med Biol. 2015;60(14):5471–96.

33. Li H, Zhu Y, Burnside ES, Huang E, Drukker K, Hoadley KA, Fan C, Conzen SD, Zuley M, Net JM, Sutton E, Whitman GJ, Morris E, Perou CM, Ji Y, Giger ML. Quantitative MRI radiomics in the prediction of molecular classifications of breast cancer subtypes in the TCGA/TCIA data set. NPJ Breast Cancer. 2016;2:16012.

34. Li H, Zhu Y, Burnside ES, Drukker K, Hoadley KA, Fan C, Conzen SD, Whitman GJ, Sutton EJ, Net JM, Ganott M, Huang E, Morris EA, Perou CM, Ji Y, Giger ML. MR imaging radiomics

273Computer-aided diagnosis with radiogenomics: analysis of the relationship between genotype…

signatures for predicting the risk of breast cancer recurrence as given by research versions of MammaPrint, Oncotype DX, and PAM50 gene assays. Radiology. 2016;281(2):382–91.

35. Guo W, Li H, Zhu Y, Lan L, Yang S, Drukker K, Morris E, Burnside E, Whitman G, Giger ML, Ji Y, Tcga Breast Phenotype Research Group. Prediction of clinical phenotypes in invasive breast carcinomas from the integration of radiomics and genom-ics data. J Med Imaging. 2015;2(4):041007.

36. Wang J, Kato F, Oyama-Manabe N, Li R, Cui Y, Tha KK, Yamashita H, Kudo K, Shirato H. Identifying triple-negative

breast cancer using background parenchymal enhancement het-erogeneity on dynamic contrast-enhanced mri: a pilot radiomics study. PLoS One. 2015;10(11):e0143308.

37. Yu J, Shi Z, Lian Y, Li Z, Liu T, Gao Y, Wang Y, Chen L, Mao Y. Noninvasive IDH1 mutation estimation based on a quan-titative radiomics approach for grade II glioma. Eur Radiol. 2017;27(8):3509–22.

38. Robert TM, Friedman TJ. The elements of statistical learn-ing, Data mining, inference and prediction. 2nd ed. New York: Springer; 2009.