Predicting the MGMT Promoter Methylation Status in T2-FLAIR

Magnetic Resonance Imaging Scans Using Machine Learning

Martyna Kurbiel

, Agata M. Wijata

and Jakub Nalepa

Department of Algorithmics and Software, Silesian University of Technology, Akademicka 16, 44-100 Gliwice, Poland

Faculty of Biomedical Engineering, Silesian University of Technology, Roosevelta 40, 41-800 Zabrze, Poland

Keywords: Glioblastoma, MGMT Promoter Methylation, MRI, Machine Learning, Radiomics.

Abstract: Glioblastoma is the most common form of brain cancer in adults, and is characterized by one of the worst

prognosis, with median survival being less than one year. Magnetic resonance imaging (MRI) plays a key

role in detecting and objectively tracking the disease by extracting quantifiable parameters of the tumor, such

as its volume or bidimensional measurements. However, it has been shown that the presence a specific genetic

sequence in a lesion, being the DNA repair enzyme O

-methylguanine-DNA methyltransferase (MGMT)

promoter methylation, may be effectively used to predict the patient’s responsiveness to chemotherapy. The

invasive process of analyzing a tissue sample to verify the MGMT promoter methylation status is time-

consuming, and may require performing multiple surgical interventions in longitudinal studies. Thus, building

non-invasive techniques of predicting the genetic subtype of glioblastoma is of utmost practical importance

to not only accelerate the overall process of determining the MGMT promoter methylation status in

glioblastoma patients, but also to minimize the number of necessary surgeries. In this paper, we tackle this

problem and propose an end-to-end machine learning classification pipeline benefitting from radiomic

features extracted from brain MRI scans, and validate it over a well-established RSNA-MICCAI Brain Tumor

Radiogenomic Classification benchmark dataset.

1 INTRODUCTION

Glioblastoma (GBM) stands out as the prevalent

malignant brain tumor among adults, and despite

extensive research spanning decades, it is still one of

the deadliest cancers, primarily attributed to its

unfavorable prognosis. Consequently, the precise

assessment of therapy response in GBM poses

significant challenges and holds immense clinical

importance (Qi et al., 2023). Although, multi-modal

magnetic resonance imaging (MRI) scans can bring

important structural information concerning such

brain lesions, their manual analysis of acquired

images is time- and cost-inefficient, it lacks

reproducibility and suffers from significant inter- and

intra-rater disagreement (Xuan et al., 2022; Hu et al.

2022). To automate the tedious process of analyzing

MRI scans, various algorithms have been emerging at

a steady pace recently. These practical challenges can

be effectively tackled by automatic brain lesions

https://orcid.org/0000-0001-6180-9979

https://orcid.org/0000-0002-4026-1569

detection and segmentation techniques. They may be

split into atlas-, image analysis-, machine learning-

based, and hybrid techniques. In the atlas-based

approaches, we exploit manually-delineated atlases to

segment unseen scans, relying on image registration

and facing challenges with diverse tumor

characteristics that are difficult to capture within an

atlas (Xing et al., 2022). Similarly, image analysis-

based algorithms, including thresholding and region-

growing techniques, are often easy to implement and

offer fast operation, but they struggle with

heterogeneous tumors and noisy images (Puttagunta

et al., 2021; Vadmal et al., 2022). Conventional

machine learning approaches offer advantages

directly related to their nature (of such methods being

data-driven), but they require heavy feature

engineering, hence elaborating manually-designed

features that would capture intrinsic brain tumor

characteristics. Finally, deep learning models

encompass a range of network architectures,

872

Kurbiel, M., Wijata, A. and Nalepa, J.

Predicting the MGMT Promoter Methylation Status in T2-FLAIR Magnetic Resonance Imaging Scans Using Machine Learning.

DOI: 10.5220/0012467400003654

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 13th International Conference on Pattern Recognition Applications and Methods (ICPRAM 2024), pages 872-879

ISBN: 978-989-758-684-2; ISSN: 2184-4313

including ensembles (Shi et al., 2021), U-Net

(Bukhari et al., 2022), encoder-decoder (Yan et al.,

2022), and more (Peiris et al., 2022) that were

thoroughly validated over the Brain Tumor

Segmentation (BraTS) Challenge throughout the

recent years, and established the current state of the

art in the field (Baid et al. 2021). Although accurate

brain lesion segmentation is of paramount importance

in order to objectively assess the tumor progression

through extracting its quantifiable characteristics,

such as its volumetric or bidimensional

measurements (Hu et al. 2022), the structural

information concerning the brain may not be enough

to fully understand the patient status and benefit from

it in planning the treatment (Beyer et al., 2020).

There have been various research efforts

indicating that the identification of a particular

genetic sequence in a lesion – specifically, the DNA

repair enzyme O

-methylguanine-DNA

methyltransferase (MGMT) promoter methylation –

can serve as an effective predictor of a patient's

responsiveness to chemotherapy (Weller et al., 2010).

Additionally, the MGMT status has become a

stratification parameter of patients with glioblastoma

within clinical trials as well. The intrusive nature of

examining a tissue sample to confirm the MGMT

promoter methylation status is time-intensive and

may necessitate multiple surgical interventions in

longitudinal studies. Consequently, the development

of non-invasive techniques for predicting the genetic

subtype of glioblastoma becomes paramount. This

not only expedites the overall process of determining

the MGMT promoter methylation status in

glioblastoma patients but also reduces the need for

multiple surgeries such patients would have to

undergo. Therefore, developing non-invasive

methods for quantifying the MGMT promoter

methylation status has been already researched in the

literature, e.g., using texture features extracted from

T2-weighted MR images and Support Vector

Machines (Korfiatis et al., 2016). It was also

demonstrated that the use of radiomic features

together with machine learning algorithms can enable

non-invasive prediction of the MGMT promoter

methylation status (Hajianfar et al., 2019) – here, a

pipeline of the radiomic feature extraction, feature

selection, and classification were employed for each

patient. Also, there are deep learning-powered

approaches, e.g., exploiting various network

architectures (Korfiatis et al., 2017). In their recent

work, Saeed et al. 2023 performed an extensive

evaluation study of an array of deep learning models

for estimating MGMT methylation status from MRI

data, and showed that the reliability of the deep

learning approaches should be verified using external

cohorts before exploiting them in clinical

applications. Here, capturing large, heterogeneous

and representative datasets that would allow for

training large-capacity learners is a practical

challenging which may ultimately hamper

generalization capabilities of deep learning models.

In this work, we tackle the problem of quantifying

the MGMT methylation status based on MRI data,

and introduced a classic machine learning algorithm

for this task. We hypothesize that the features

extracted from the whole brain region scanned using

the T2 Fluid Attenuation Inversion Recovery (T2-

FLAIR) MR sequence, as such sequences have been

designed to suppress the signal from cerebrospinal

fluid, providing improved visualization of lesions

near cerebrospinal fluid spaces, may be utilized in

differentiating the MGMT methylation status (Alpar,

2023). Here, since the lesion segmentation step is

skipped in our processing chain, we may not only

accelerate the computation, as a single MR sequence

is processed, but we can also rely on the widely-

established brain extraction (skull stripping)

algorithms (Isensee et al., 2019) for removing the

skull that are known to be generalizing well over the

unseen MR scans. Once the T2-FLAIR sequence is

skull-stripped, we extract nearly 120 radiomic-based

features that are fed (with or without additional

dimensionality reduction) to the classification engine.

The generalization capabilities of the proposed

technique for quantifying the MGMT methylation

status were verified over the RSNA-MICCAI Brain

Tumor Radiogenomic Classification benchmark

dataset (Baid et al. 2021; Bakas et al., 2017a; Bakas

et al., 2017b; Bakas et al., 2017c; Menze et al., 2015).

In this study, we frame the problem of assessing the

MGMT methylation status as the classification task,

with the patients being assigned to unmethylated and

methylated classes.

The remainder of the paper is structured as

follows. In Section 2, we present the RSNA-MICCAI

Brain Tumor Radiogenomic Classification

benchmark dataset, and introduce our machine

learning pipeline for assessing the MGMT

methylation status based on the radiomic-based

features extracted from T2-FLAIR MR sequences. In

Section 3, we report and discuss the experimental

study performed to investigate the generalization

capabilities of the algorithms, as well as to verify the

impact of various dimensionality reduction

techniques on its capabilities (both classic and deep

learning-powered, with the latter benefiting from

autoencoder architectures). Finally, Section 4

summarized the findings and sheds more light on the

Predicting the MGMT Promoter Methylation Status in T2-FLAIR Magnetic Resonance Imaging Scans Using Machine Learning

873

most promising research directions that may emerge

from the results obtained in this article.

a) The axial plane

) The sagittal plane

)

The coronal

lane

Figure 1: An example of a skull-stripped T2-FLAIR MR

frames (visualized in the false-color scheme), together with

the corresponding brain regions in the a) axial, b) sagittal,

and c) coronal planes.

2 MATERIALS AND METHODS

In this section, we discuss the dataset used in our

study (Section 2.1). In Section 2.2, we present the

most important steps of our processing chain for

classifying the patients into the unmethylated and

methylated classes, based on the radiomic features

extracted from T2-FLAIR MR sequences.

2.1 The RSNA-ASNR-MICCAI Brain

Tumor Segmentation Dataset

In this study, we build upon the RSNA-ASNR-

MICCAI Brain Tumor Segmentation (BraTS)

benchmark dataset (the 2021 edition, for which the

clinical information related to the MGMT promoter

methylation status was obtained as well) (Baid et al.

2021; Bakas et al., 2017a; Bakas et al., 2017b; Bakas

et al., 2017c; Menze et al., 2015). This dataset

contains multi-modal MRI scans captured with

different protocols and scanners from multiple

institutions, and the BraTS dataset is commonly

considered the state-of-the-art benchmark dataset for

confronting the brain tumor segmentation algorithms,

thanks to its size and heterogeneity. The MRI scans

contained within the dataset were interpolated to the

same shape (the size of an MRI scan is 240 × 240 ×

155, therefore there are 155 images of 240 × 240 MR

images, with the voxel size of 1 mm

). All of the

available images are skull-stripped – a set of example

T2-FLAIR frames (obtained for a single patient) with

the corresponding brain ground-truth segmentation

masks are rendered in Figure 1.

Figure 2: Distribution of the unmethylated and methylated

patients in the dataset used in this study.

The MGMT promoter methylation status data was

defined as a binary label, corresponding to the

unmethylated and methylated patients. The

distribution of the methylated and unmethylated

patients within the training set of BraTS 2021 (for

which the ground-truth labels are known, as they were

revealed by the organizers of the challenge) is

visualized in Figure 2. Out of all 585 patients, we

removed nine patients due to an incorrect registration

of their brain segmentation masks and corresponding

image data. Therefore, the final dataset included 576

ICPRAM 2024 - 13th International Conference on Pattern Recognition Applications and Methods

874

patients with the MRI scans and the corresponding

MGMT promoter methylation status. We can observe

that the dataset is balanced, and includes a similar

number of unmethylated and methylated patients.

2.2 Predicting the MGMT Promoter

Methylation Status Using Machine

Learning and Radiomic Features

In this work, we introduce an end-to-end processing

chain benefiting from classic machine learning

classification models (trained in a supervised way)

operating over the radiomic features extracted from

T2-FLAIR sequences of brain MRI (Figure 3). The

feature extraction may be followed by an optional

dimensionality reduction step which can play a

pivotal role if a very large number of radiomic

features are extracted, as it may easily lead to

overfitting the model to the training data (Kotowski

et al., 2023). Of note, our approach for determining

the MGMT promoter methylation status offers a high

level of flexibility, and the specific algorithms may be

easily updated at each processing step – this

flexibility will be further proven in the experimental

section of this article.

Figure 3: A high-level flowchart presenting the proposed

processing chain. The optional step is rendered as a dashed

block, whereas the input and output steps are presented as

white and orange ones.

The input T2-FLAIR images undergo brain

extraction, which might be performed using an array

of thoroughly-evaluated state-of-the-art techniques,

such as the HD-BET algorithm (Isensee et al., 2019)

(note that the scans included in BraTS are already

skull-stripped, hence this step was omitted in our

study). Afterwards, we extract the following radiomic

features (as suggested by van Griethuysen et al., 2017

and by Ponikiewski et al., 2022) from the 3D brain

region of the T2-FLAIR scan:

• First Order Statistics (18 features),

• Shape-based (3D) features (14 features),

• Gray Level Co-occurrence Matrix (24 features)

• Gray Level Run Length Matrix (16 features),

• Gray Level Size Zone Matrix (16 features),

• Neighboring Gray Tone Difference Matrix (5

features),

• Gray Level Dependence Matrix (14 features).

The majority of the features are in compliance

with the feature definitions as suggested by the

Imaging Biomarker Standardization Initiative

(Zwanenburg et al., 2020). Overall, we extract 119

features (which were scaled to the unit variance).

Since the number of features is large, especially

when confronted with a relatively small number of

patients, exploiting all of them while training

supervised learners may easily lead to overfitting

them to the training set, hence memorizing it – it

would render them impossible to generalize over the

unseen test patients (Ying et al., 2019). To deal with

this issue, we exploit the additional (yet optional)

dimensionality reduction step, and employ the

following techniques for this task (although we are

aware that the hyperparameters of the following data

dimensionality methods are tunable, we present them

here, rather than in the experimental section in order

to make this section self-contained):

• Principal component analysis (PCA), for which

the number of principal components (PCs) was

selected to explain 98% of the data variance (21

PC were exploited). In Figure 4, we can observe

that exploiting just two PCs would make the

classification process (i.e., distinguishing the

methylated and unmethylated patients) virtually

impossible due to heavy overlaps across these

two classes in the PC space for 2 PCs.

Figure 4: The first two PCs show that discriminating

unmethylated (black dots) and methylated (orange dots)

patients would be virtually impossible using only two PCs.

In this study, we selected 21 PCs to explain 98% variance

within the dataset.

• Autoencoder (AE) with a fully-connected

architecture with the scaled exponential linear

unit activations, containing two encoding and

decoding layers (with 50 and 30 neurons), and

Predicting the MGMT Promoter Methylation Status in T2-FLAIR Magnetic Resonance Imaging Scans Using Machine Learning

875

elaborating the latent representation of 21

features (to ensure consistency with the number

of PCs elaborated by PCA).

• Feature selection (FS), where we selected 21

features with the largest variance (as previously,

we ensured consistency with the number of PCs).

Such variance-based feature selection might be

useful to ensure interpretability of the extracted

features (this is not necessarily the case for the

radiomic features, as they may be fundamentally

challenging to interpret by human readers).

In Figure 5, we render the distributions of the

selected features for all dimensionality reduction

techniques – these features (extracted by each

dimensionality reduction approach) are later fed into

the supervised learner for elaborating the predicted

class label (i.e., methylated or unmethylated patient).

There are numerous established supervised

classification models that could be exploited in our

processing pipeline. In this study, we investigated the

following machine learning models which have

proven their classification capabilities in a range of

real-world applications: logistic regression (LR),

support vector machines (SVMs), random forests

(RFs), k-nearest neighbor classifiers (k-NN), extreme

gradient boosting classifiers (XGBoost), and artificial

neural networks (ANNs) with a single hidden layer

containing 10 neurons. As for the feature extraction

and dimensionality reduction techniques, other

machine learning models (also deep learning

techniques) can be easily exploited in our approach.

3 EXPERIMENTAL STUDY

In this section, we discuss the results obtained in our

experimental study. To quantify the generalization

capabilities of the classification engine, we follow the

5-fold cross-validation procedure, where each fold is

stratified according to the ratio of unmethylated and

methylated patients within the full dataset. The

performance of the models was assessed using classic

metrics, including precision (Pr), recall (Re), F1 score

and the Matthews's correlation coefficient (MCC).

All metrics should be maximized, where one

indicates the perfect classification (additionally, we

tracked accuracy during the ANN training to verify if

it started overfitting). The hyperparameters of all

investigated machine learning models were fine-

tuned using an internal cross-validation procedure

performed over the corresponding training set (the

test set in the k-fold cross-validation approach was

never used here).

a) Principal component analysis

b) Autoencoder

c) Feature selection

Figure 5: Distribution of the features selected using a)

principal component analysis, b) a fully-connected

autoencoder, and c) variance-based feature selection.

Finally, to make sure that the processing chain is

straightforward to reproduce (the full approach was

implemented in Python 3.6), we exploited a well-

established pyradiomics package to extract radiomic

features from the brain areas, and the scikit-learn

package for the classification models.

In Figure 6, we gather the experimental results

(quantified as all above-mentioned quality metrics)

obtained for all investigated machine learning models

and dimensionality reduction techniques, averaged

across all five test folds. We can appreciate that

various dimensionality reduction gave consistently

similar results for virtually all classification models

ICPRAM 2024 - 13th International Conference on Pattern Recognition Applications and Methods

876

Figure 6: Classification results (averaged across all five test folds) obtained for all investigated machine

learning classification models and dimensionality reduction techniques (the black color corresponds to

principal component analysis, red to the autoencoder, and yellow to variance-based feature selection).

Table 1: The results obtained using the ANN model without

and with regularization techniques applied (averaged across

all test sets in the five-fold cross-validation scenario). The

best metrics for each dimensionality reduction approach are

boldfaced.

Regularization Metric PCA AE FS

None F1 0.53 0.54 0.51

0.53 0.54 0.54

Re 0.55 0.57 0.49

MCC 0.01 0.05 0.03

Dropout F1 0.55 0.49 0.57

0.54 0.56 0.58

Re 0.56 0.43 0.56

MCC 0.04 0.06 0.12

Dropout and

early stopping

F1 0.56 0.56 0.56

0.59 0.59 0.59

Re 0.52 0.54 0.54

MCC 0.13 0.12 0.12

(the smallest differences between different

dimensionality reduction routines were captured for

the LR classifier), with PCA outperforming the other

methods for RF. Here, this model resulted in the

highest recall values which is of paramount clinical

significance, as identifying methylated patients may

lead to designing their more effective treatment

pathways. Of note, it was observable that the ANN

model started overfitting the training set – as an

example for the PCA dimensionality reduction, the

accuracy over the training folds exceeded 0.9, with

the corresponding accuracy over the validation set

reaching approx. 0.6. This phenomenon was,

however, observed for all dimensionality reduction

approaches, indicating that the training sample may

be too small to elaborate a well-generalizing

classifiers. To verify if applying additional

regularization techniques could help improve the

abilities of the ANN model, we investigated two

additional (yet well-established in the field)

regularization approaches, being the dropout within

the ANN, together with an early stopping routine. The

results gathered in Table 1 indeed confirm that

applying additional regularization techniques help

improve the generalization capabilities of the ANN

models.

4 CONCLUSIONS

Glioblastoma is the most common form of brain

cancer, and the detailed profiling of patients suffering

from this disease is of pivotal importance. We

approached this issue, and proposed a machine

learning pipeline to predict the MGMT promoter

methylation from T2-FLAIR, as it is an important

biomarker for the patient prognosis. The experiments

indicated that radiomic features extracted from

whole-brain scans allow to elaborate classifiers that

identify the methylated patients. The generalization

of models, thus their clinical utility might be

improved by gathering more heterogeneous and

representative training sets, as we observed that the

models started overfitting, and by explicitly tackling

the problem of the dataset imbalance. This issue may

be also tackled using model-level regularization

which was shown effective in this study.

Predicting the MGMT Promoter Methylation Status in T2-FLAIR Magnetic Resonance Imaging Scans Using Machine Learning

877

ACKNOWLEDGEMENTS

This work was supported by the Silesian University

of Technology funds through the grant for

maintaining and developing research potential, and

by the Silesian University of Technology funds

through the Excellence Initiative—Research

University program (Grant 02/080/SDU/10-21-01).

REFERENCES

Alpar, O. (2023). A mathematical fuzzy fusion framework

for whole tumor segmentation in multimodal MRI

using Nakagami imaging. Expert Systems with

Applications, 216. https://doi.org/10.1016/j.eswa.

2022.119462.

Baid, U., Ghodasara, S., Bilello, M., Mohan, S., Calabrese,

E., Colak, E., Farahani, K., Kalpathy-Cramer, J.,

Kitamura, F.C., Pati, S., Prevedello, L.M., Rudie, J.D.,

Sako, C., Shinohara, R.T., Bergquist, T., Chai, R.,

Eddy, J.A., Elliott, J., Reade, W.C., Schaffter, T., Yu,

T., Zheng, J., Annotators, B., Davatzikos, C., Mongan,

J.T., Hess, C.P., Cha, S., Villanueva-Meyer, J.E.,

Freymann, J.B., Kirby, J.S., Wiestler, B., Crivellaro,

P.S., R.Colen, R., Kotrotsou, A., Marcus, D.,

Milchenko, M., Nazeri, A., Fathallah-Shaykh, H.M.,

Wiest, R., Jakab, A., Weber, M., Mahajan, A., Menze,

B.H., Flanders, A.E., & Bakas, S. (2021). The RSNA-

ASNR-MICCAI BraTS 2021 Benchmark on Brain

Tumor Segmentation and Radiogenomic Classification.

ArXiv, abs/2107.02314.

Bakas, S., Akbari, H., Sotiras, A., Bilello, M., Rozycki, M.,

Kirby, J. S., Freymann, J. B., Farahani, K., &

Davatzikos, C. (2017). Advancing The Cancer Genome

Atlas glioma MRI collections with expert segmentation

labels and radiomic features. Scientific data, 4, 170117.

https://doi.org/10.1038/sdata.2017.117.

Bakas S, Akbari H, Sotiras A, Bilello M, Rozycki M, Kirby

J, Freymann J, Farahani K, Davatzikos C. (2017).

Segmentation Labels for the Pre-operative Scans of the

TCGA-GBM collection [Data set]. The Cancer Imaging

Archive. DOI: 10.7937/K9/TCIA.2017.KLXWJJ1Q.

Bakas S, Akbari H, Sotiras A, Bilello M, Rozycki M, Kirby

J, Freymann J, Farahani K, Davatzikos C. (2017)

Advancing The Cancer Genome Atlas glioma MRI

collections with expert segmentation labels and

radiomic features Nature Scientific Data, 4:170117

DOI: 10.1038/sdata.2017.117.

Beyer, T., Bidaut, L., Dickson, J., Kachelriess, M.,

Kiessling, F., Leitgeb, R., Ma, J., Shiyam Sundar, L. K.,

Theek, B., & Mawlawi, O. (2020). What scans we will

read: imaging instrumentation trends in clinical

oncology. Cancer imaging : the official publication of

the International Cancer Imaging Society, 20(1), 38.

https://doi.org/10.1186/s40644-020-00312-3

Bukhari, S.T., Mohy-ud-Din, H. (2022). E1D3 U-Net for

Brain Tumor Segmentation: Submission to the RSNA-

ASNR-MICCAI BraTS 2021 challenge. In: Crimi, A.,

Bakas, S. (eds) Brainlesion: Glioma, Multiple

Sclerosis, Stroke and Traumatic Brain Injuries.

BrainLes 2021. Lecture Notes in Computer Science, vol

12963. Springer, Cham. https://doi.org/10.1007/978-3-

031-09002-8_25

van Griethuysen, J. J. M., Fedorov, A., Parmar, C., Hosny,

A., Aucoin, N., Narayan, V., Beets-Tan, R. G. H.,

Fillion-Robin, J. C., Pieper, S., & Aerts, H. J. W. L.

(2017). Computational Radiomics System to Decode

the Radiographic Phenotype. Cancer research, 77(21),

e104–e107. https://doi.org/10.1158/0008-5472.CAN-

17-0339

Hajianfar, G., Shiri, I., Maleki, H., Oveisi, N., Haghparast,

A., Abdollahi, H., Oveisi, M. (2019). Noninvasive O6

Methylguanine-DNA Methyltransferase Status

Prediction in Glioblastoma Multiforme Cancer Using

Magnetic Resonance Imaging Radiomics Features:

Univariate and Multivariate Radiogenomics Analysis.

World Neurosurgery. DOI: 10.1016/j.wneu.2019.

08.232.

Hu, X., Luo, W., Hu, J., Guo, S., Huang, W., Scott, M. R.,

Wiest, R., Dahlweid, M., & Reyes, M. (2020). Brain

SegNet: 3D local refinement network for brain lesion

segmentation. BMC medical imaging, 20(1), 17.

https://doi.org/10.1186/s12880-020-0409-2.

Korfiatis, P., Kline, T. L., Coufalova, L., Lachance, D. H.,

Parney, I. F., Carter, R. E., Buckner, J. C., & Erickson,

B. J. (2016). MRI texture features as biomarkers to

predict MGMT methylation status in glioblastomas.

Medical physics, 43(6), 2835–2844. https://doi.org/

10.1118/1.4948668.

Korfiatis, P., Kline, T. L., Lachance, D. H., Parney, I. F.,

Buckner, J. C., & Erickson, B. J. (2017). Residual Deep

Convolutional Neural Network Predicts MGMT

Methylation Status. Journal of digital imaging, 30(5),

622–628. https://doi.org/10.1007/s10278-017-0009-z.

Kotowski, K., Kucharski, D., Machura, B., Adamski, S.,

Gutierrez Becker, B., Krason, A., Zarudzki, L., Tessier,

J., & Nalepa, J. (2023). Detecting liver cirrhosis in

computed tomography scans using clinically-inspired

and radiomic features. Computers in biology and

medicine, 152, 106378. https://doi.org/10.1016/j.

compbiomed.2022.106378.

Isensee, F., Schell, M., Pflueger, I., Brugnara, G.,

Bonekamp, D., Neuberger, U., Wick, A., Schlemmer,

H. P., Heiland, S., Wick, W., Bendszus, M., Maier-

Hein, K. H., & Kickingereder, P. (2019). Automated

brain extraction of multisequence MRI using artificial

neural networks. Human brain mapping, 40(17), 4952–

4964. https://doi.org/10.1002/hbm.24750.

Menze, B. H., Jakab, A., Bauer, S., Kalpathy-Cramer, J.,

Farahani, K., Kirby, J., Burren, Y., Porz, N., Slotboom,

J., Wiest, R., Lanczi, L., Gerstner, E., Weber, M. A.,

Arbel, T., Avants, B. B., Ayache, N., Buendia, P.,

Collins, D. L., Cordier, N., Corso, J. J., … Van

Leemput, K. (2015). The Multimodal Brain Tumor

Image Segmentation Benchmark (BRATS). IEEE

transactions on medical imaging, 34(10), 1993–2024.

https://doi.org/10.1109/TMI.2014.2377694.

ICPRAM 2024 - 13th International Conference on Pattern Recognition Applications and Methods

878

Peiris, H., Chen, Z., Egan, G., Harandi, M. (2022).

Reciprocal Adversarial Learning for Brain Tumor

Segmentation: A Solution to BraTS Challenge 2021

Segmentation Task. In: Crimi, A., Bakas, S. (eds)

Brainlesion: Glioma, Multiple Sclerosis, Stroke and

Traumatic Brain Injuries. BrainLes 2021. Lecture

Notes in Computer Science, vol 12962. Springer,

Cham. https://doi.org/10.1007/978-3-031-08999-2_13.

Ponikiewski, W., Nalepa, J. (2022). Deep Learning Meets

Radiomics For End-To-End Brain Tumor MRI

Analysis. 1301-1305. 10.1109/ICIP46576.2022.

9897847.

Puttagunta, M., & Ravi, S. (2021). Medical image analysis

based on deep learning approach. Multimedia tools and

applications, 80(16), 24365–24398. https://doi.org/10.

1007/s11042-021-10707-4.

Qi, D., Li, J., Quarles, C. C., Fonkem, E., & Wu, E. (2023).

Assessment and prediction of glioblastoma therapy

response: challenges and opportunities. Brain : a

journal of neurology, 146(4), 1281–1298.

https://doi.org/10.1093/brain/awac450

Saeed, N., Ridzuan, M., Alasmawi, H., Sobirov, I., Yaqub,

M. (2023). MGMT promoter methylation status

prediction using MRI scans? An extensive

experimental evaluation of deep learning models.

Medical Image Analysis. 90. 102989.

10.1016/j.media.2023.102989.

Shi, Y., Micklisch, C., Mushtaq, E., Avestimehr, S., Yan,

Y., Zhang, X. (2022). An Ensemble Approach to

Automatic Brain Tumor Segmentation. In: Crimi, A.,

Bakas, S. (eds) Brainlesion: Glioma, Multiple

Sclerosis, Stroke and Traumatic Brain Injuries.

BrainLes 2021. Lecture Notes in Computer Science, vol

12963. Springer, Cham. https://doi.org/10.1007/978-3-

031-09002-8_13.

Vadmal, V., Junno, G., Badve, C., Huang, W., Waite, K.

A., & Barnholtz-Sloan, J. S. (2020). MRI image

analysis methods and applications: an algorithmic

perspective using brain tumors as an exemplar. Neuro-

oncology advances, 2(1), vdaa049. https://doi.org/10.

1093/noajnl/vdaa049.

Weller, M., Stupp, R., Reifenberger, G., Brandes, A. A.,

van den Bent, M. J., Wick, W., & Hegi, M. E. (2010).

MGMT promoter methylation in malignant gliomas:

ready for personalized medicine?. Nature reviews.

Neurology, 6(1), 39–51. https://doi.org/10.1038/

nrneurol.2009.197.

Xing, F., Liu, X., Kuo, C. J., Fakhri, G. E., & Woo, J.

(2022). Brain MR Atlas Construction Using Symmetric

Deep Neural Inpainting. IEEE journal of biomedical

and health informatics, 26(7), 3185–3196.

https://doi.org/10.1109/JBHI.2022.3149754.

Xuan, K., Xiang, L., Huang, X., Zhang, L., Liao, S., Shen,

D., & Wang, Q. (2022). Multimodal MRI

Reconstruction Assisted With Spatial Alignment Network.

IEEE transactions on medical imaging, 41(9), 2499–

2509. https://doi.org/10.1109/TMI.2022.3164050.

Yan, B., Wei, Y., Jagtap, J., Moassefi, M., Vera Gracia,

DV., & Singh, Y., Vahdati, S., Faghani, S., Erickson,

B., Conte, G. (2022). MRI Brain Tumor Segmentation

Using Deep Encoder-Decoder Convolutional Neural

Networks. In: Crimi, A., Bakas, S. (eds) Brainlesion:

Glioma, Multiple Sclerosis, Stroke and Traumatic

Brain Injuries. BrainLes 2021. Lecture Notes in

Computer Science, vol 12963. Springer, Cham.

https://doi.org/10.1007/978-3-031-09002-8_7.

Ying, X. (2019). An Overview of Overfitting and its

Solutions. Journal of Physics: Conference Series. 1168.

022022. 10.1088/1742-6596/1168/2/022022.

Zwanenburg, A., Vallières, M., Abdalah, M. A., Aerts, H.

J. W. L., Andrearczyk, V., Apte, A., Ashrafinia, S.,

Bakas, S., Beukinga, R. J., Boellaard, R., Bogowicz,

M., Boldrini, L., Buvat, I., Cook, G. J. R., Davatzikos,

C., Depeursinge, A., Desseroit, M. C., Dinapoli, N.,

Dinh, C. V., Echegaray, S., … Löck, S. (2020). The

Image Biomarker Standardization Initiative:

Standardized Quantitative Radiomics for High-

Throughput Image-based Phenotyping. Radiology,

295(2), 328–338. https://doi.org/10.1148/radiol.

2020191145.

Predicting the MGMT Promoter Methylation Status in T2-FLAIR Magnetic Resonance Imaging Scans Using Machine Learning

879