Improving brain tumor characterization on MRI by probabilistic neural networks and non-linear transformation of textural features

doi:10.1016/j.cmpb.2007.10.007

Computer Methods and Programs in Biomedicine

Volume 89, Issue 1, January 2008, Pages 24-32

https://doi.org/10.1016/j.cmpb.2007.10.007 Get rights and content

Abstract

The aim of the present study was to design, implement and evaluate a software system for discriminating between metastatic and primary brain tumors (gliomas and meningiomas) on MRI, employing textural features from routinely taken T1 post-contrast images. The proposed classifier is a modified probabilistic neural network (PNN), incorporating a non-linear least squares features transformation (LSFT) into the PNN classifier. Thirty-six textural features were extracted from each one of 67 T1-weighted post-contrast MR images (21 metastases, 19 meningiomas and 27 gliomas). LSFT enhanced the performance of the PNN, achieving classification accuracies of 95.24% for discriminating between metastatic and primary tumors and 93.48% for distinguishing gliomas from meningiomas. To improve the generalization of the proposed classification system, the external cross-validation method was also used, resulting in 71.43% and 81.25% accuracies in distinguishing metastatic from primary tumors and gliomas from meningiomas, respectively. LSFT improved PNN performance, increased class separability and resulted in dimensionality reduction.

Introduction

According to a recent statistical report published by the Central Brain Tumor Registry of the United States (CBTRUS), approximately 39,550 people were newly diagnosed with primary benign and primary malignant brain tumors [1], [2] in 2002 [3]. Furthermore, in 2000, more than 81,000 people, in the United States alone, were living with a primary malignant brain tumor and 267,000 were living with a primary benign brain tumor. The same report indicates that the incidence rate of primary brain tumors, whether benign or malignant, is 14 per 100,000, while median age at diagnosis is 57 years [3].

Secondary or metastatic brain tumors [1], in contrast to primary brain tumors, originate in tissues outside the central nervous system and are a common complication of systemic cancer. Brain metastases outnumber primary brain tumors and are currently classified as the most frequent intracranial tumors. Other studies indicate that brain metastases occur in 20–40% of all cancer patients and that more than 100,000 individuals per year will develop brain metastases [3].

Today, imaging techniques, like magnetic resonance imaging (MRI), are used to locate the position and extent of brain tumors. MRI can provide information about brain tissues, from a variety of excitation sequences. Compared with other diagnostic imaging modalities, such as computerized tomography, MRI provides superior contrast for different brain tissues [4]. Additionally, MR images encapsulate valuable information regarding numerous tissue parameters (proton density, spin–lattice (T1) and spin–spin (T2) relaxation times, flow velocity and chemical shift), which lead to more accurate brain tissue characterization. These unique advantages have characterized MRI as the method of choice in brain tumor studies [5].

Brain tumor characterization is a process that requires a complicated assessment of the various MR image features and is typically performed by experienced radiologists. An expert radiologist performs this task with a significant degree of precision and accuracy, despite the inherently subjective nature of many of the decisions associated with this process. Nevertheless, in the effort to deliver more effective treatment, clinicians are continuously seeking for greater accuracy in the pathological characterization of brain tissues from imaging investigations [6].

Section snippets

Background and design considerations

To this need, image analysis techniques have been employed in previous studies for the extraction of diagnostic information from MR images [6], [7], [8]. These studies have employed pattern recognition and texture analysis techniques to characterize human brain tumors. In a recent study [9], an SVM-based classification system has achieved 95% overall accuracy in discriminating between gliomas and meningiomas. In another study [7], the hierarchical ascending classification with correspondence

Data acquisition

A total number of 67 T1-weighted gadolinium-enhanced MR images were obtained from the Hellenic Airforce Hospital with verified intracranial tumors, using a SIEMENS-Sonata 1.5 Tesla MR Unit. The image dataset comprised 21 metastases, 19 meningiomas and 27 gliomas. From each case, only T1-weighted post-contrast (Gadolinium) images, with spin echo (SE) sequence, echo time (TE = 15 ms) and repetition time (TR = 500 ms), were used for further analysis. The reason for employing T1 post-contrast images is

Experimental results

Fig. 3 shows a scatter diagram of the classes involved. The complexity of the problem has led us to adopt a hierarchical decision tree structure (see Fig. 2). The overall classification accuracy at the first level of the decision tree was 94.03% employing the cubic LSFT-PNN classifier. Individual accuracies in discriminating between primary and secondary brain tumors were 93.48% and 95.24%, respectively (Table 1). Best feature vector, used for the optimal design of the cubic LSFT-PNN

Discussion

The LSFT-PNN and the PNN classification schemes were optimized with respect to parameter settings and available feature data. The spread of Gaussian function for the LSFT-PNN and the PNN classifiers was experimentally set equal to σ = 0.3.

In accordance with our findings, the LSFT-PNN outperformed the PNN at both levels of the decision tree. At the first level, the LSFT-PNN achieved a sensitivity of 93.48% against PNN's 86.96% in correctly characterizing primary tumors, assigning three more

Acknowledgement

Funding by the University of Patras Research Committee under the basic research program “K. Karatheodori”, project title “Computer Assisted Diagnosis of Brain Tumors based on Statistical Methods and Pattern Recognition Techniques” is gratefully acknowledged.

References (26)

L.S. Ashby et al.
Central nervous system tumors
Update Cancer Therapeut.
(2006)
G. Cruickshank
Tumours of the brain
Surgery (Oxford)
(2004)
N.D. Doolittle
State of the science in brain tumor classification
Semin. Oncol. Nurs.
(2004)
R.A. Lerski et al.
MR image texture analysis—an approach to tissue characterization
Magn. Reson. Imaging
(1993)
S. Herlidou-Meme et al.
MRI texture analysis on texture test objects, normal brain and intracranial tumors
Magn. Reson. Imaging
(2003)
L.R. Schad et al.
MR tissue characterization of intracranial tumors by means of texture analysis
Magn. Reson. Imaging
(1993)
A. Devos et al.
The use of multivariate MR imaging intensities versus metabolic data from MR spectroscopic imaging for brain tumour classification
J. Magn. Reson.
(2005)
G.-Z. Li et al.
Degree prediction of malignancy in brain glioma using support vector machines
Comput. Biol. Med.
(2006)
L. Lukas et al.
Brain tumor classification based on long echo proton MRS signals
Artif. Intell. Med.
(2004)
A. Devos et al.
Classification of brain tumours using short echo time 1H MR spectra
J. Magn. Reson.
(2004)

Y.-D. Cho et al.

1H-MRS metabolic patterns for distinguishing between meningiomas and other brain tumors

Magn. Reson. Imaging

(2003)

M.M. Galloway

Texture analysis using grey level run lengths

Comp. Graph. Image Proc.

(1975)

D.F. Specht

Probabilistic neural networks

Neural Networks

(1990)

Cited by (136)

Performance of convolutional neural networks for the classification of brain tumors using magnetic resonance imaging
2024, Heliyon
Brain tumors are a diverse group of neoplasms that are challenging to detect and classify due to their varying characteristics. Deep learning techniques have proven to be effective in tumor classification. However, there is a lack of studies that compare these techniques using a common methodology. This work aims to analyze the performance of convolutional neural networks in the classification of brain tumors. We propose a network consisting of a few convolutional layers, batch normalization, and max-pooling. Then, we explore recent deep architectures, such as VGG, ResNet, EfficientNet, or ConvNeXt. The study relies on two magnetic resonance imaging datasets with over 3000 images of three types of tumors –gliomas, meningiomas, and pituitary tumors–, as well as images without tumors. We determine the optimal hyperparameters of the networks using the training and validation sets. The training and test sets are used to assess the performance of the models from different perspectives, including training from scratch, data augmentation, transfer learning, and fine-tuning. The experiments are performed using the TensorFlow and Keras libraries in Python. We compare the accuracy of the models and analyze their complexity based on the capacity of the networks, their training times, and image throughput. Several networks achieve high accuracy rates on both datasets, with the best model achieving 98.7% accuracy, which is on par with state-of-the-art methods. The average precision for each type of tumor is 94.3% for gliomas, 93.8% for meningiomas, 97.9% for pituitary tumors, and 95.3% for images without tumors. VGG is the largest model with over 171 million parameters, whereas MobileNet and EfficientNetB0 are the smallest ones with 3.2 and 5.9 million parameters, respectively. These two neural networks are also the fastest to train with 23.7 and 25.4 seconds per epoch, respectively. On the other hand, ConvNext is the slowest model with 58.2 seconds per epoch. Our custom model obtained the highest image throughput with 234.37 images per second, followed by MobileNet with 226 images per second. ConvNext obtained the smallest throughput with 97.35 images per second. ResNet, MobileNet, and EfficientNet are the most accurate networks, with MobileNet and EfficientNet demonstrating superior performance in terms of complexity. Most models achieve the best accuracy using transfer learning followed by a fine-tuning step. However, data augmentation does not contribute to increasing the accuracy of the models in general.
A computer-aided grading of glioma tumor using deep residual networks fusion
2022, Computer Methods and Programs in Biomedicine
Background and objectives: Among different cancer types, glioma is considered as a potentially fatal brain cancer that arises from glial cells. Early diagnosis of glioma helps the physician in offering effective treatment to the patients. Magnetic Resonance Imaging (MRI)-based Computer-Aided Diagnosis for the brain tumors has attracted a lot of attention in the literature in recent years. In this paper, we propose a novel deep learning-based computer-aided diagnosis for glioma tumors.
Methods: The proposed method incorporates a two-level classification of gliomas. Firstly, the tumor is classified into low-or high-grade and secondly, the low-grade tumors are classified into two types based on the presence of chromosome arms $1 p / 19 q$ . The feature representations of four residual networks have been used to solve the problem by utilizing transfer learning approach. Furthermore, we have fused these trained models using a novel Dempster-shafer Theory (DST)-based fusion scheme in order to enhance the classification performance. Extensive data augmentation strategies are also utilized to avoid over-fitting of the discrimination models.
Results: Extensive experiments have been performed on an MRI dataset to show the effectiveness of the method. It has been found that our method achieves $95.87 %$ accuracy for glioma classification. The results obtained by our method have also been compared with different existing methods. The comparative study reveals that our method not only outperforms traditional machine learning-based methods, but it also produces better results to state-of-the-art deep learning-based methods.
Conclusion: The fusion of different residual networks enhances the tumor classification performance. The experimental findings indicates that Dempster-shafer Theory (DST)-based fusion technique produces superior performance in comparison to other fusion schemes.
Classification and Pathologic Diagnosis of Gliomas in MR Brain Images
2022, Procedia Computer Science
Detection and classification of a brain tumour in medical images is always a challenging task, as the treatment may lead to Radiosurgery depending upon the exact shape, size, and position of a tumour. To provide a complete characterization of glioma and the degree of malignancy, classification and grading are vital. In some complicated cases, localization of the tumour, comparison of tumour tissues with adjacent regions, to make it clear for detection, and finally classification without human intervention is need of the hour. Out of the available numerous imaging techniques to detect and classify brain tumours, MRI is the most suitable non-invasive technique and has superior image quality. The edge over other techniques is its superior soft-tissue resolution and the ability to acquire different images employing contrast-enhanced agents. This research work classifies Low-Grade Glioma (LGG) and High-Grade Glioma(HGG) tumours on the basis of features obtained from extracted tumor region. The results of the classification of HGG and LGG tumours using the most prominent features are validated by five-fold cross-validation. Satisfactory and encouraging classification results are obtained using large and publicly available clinical datasets. Using the extracted features, a computer-assisted algorithm is developed that uses a machine-learning algorithm along with Gradient-Based Kernel Selection Graph Cut to provide binary classification with an accuracy of 94.6% with T1ce sequence of MRI. The proposed models can be employed to assist physicians and radiologists in the early pathological detection and classification of gliomas. The proposed framework also can be a model for validating brain tumours and their initial screening for their grading classification.
High Grade Brain Cancer Segmentation by means of Deep Learning
2022, Procedia Computer Science
Image segmentation is becoming a component of medical image processing with a significant role to analyze gross anatomy, to locate an infirmity and to plan the surgical procedures. Segmentation of brain magnetic resonance Imaging (MRI) is of increasing importance for the accurate diagnosis. For this reason, precise and accurate segmentation of brain MRI is a challenging task. In this paper we propose an approach for brain segmentation from MRI. Our method relies on a modified version of the U-Net convolutional neural network. Experiments performed on high grade brain cancer MRI demonstrate the effectiveness of the proposed approach for high grade brain cancer segmentation.
Deep autoencoder-based automated brain tumor detection from MRI data
2022, Artificial Intelligence-Based Brain-Computer Interface
Brain tumor detection from magnetic resonance (MR) image samples is the core way for radiologists, specialists, and physicians. Artificial intelligence-based MR image classification can make a great contribution to the decision-making process of physicians due to both healthcare personnel shortages and workload. Deep learning approaches are frequently used since they have high performance in medical image classification tasks. In this study, a novel and effective method based on a deep autoencoder was proposed for brain tumor detection from MR image samples. In the deep autoencoder structure, convolutional layers were used instead of dense layers. The deep feature sets were obtained from the last encoded layer of the deep autoencoder model. The deep features were reduced with the variance threshold algorithm. A series of classifiers, which are composed of support vector machine (SVM), decision trees (DT), k-nearest neighbor (KNN), and ensembles, were utilized in the classification process. The best classification metrics were provided with the SVM classifier using the radial basis function (RBF) kernel. The achievement of the proposed method was compared to both the existing approaches using the same dataset and the deep learning models such as VGG16 and AlexNet models.
Low-grade magnetic resonance image enhancement using adaptive sigmoid transformation function
2024, Health and Technology

View all citing articles on Scopus

View full text

Improving brain tumor characterization on MRI by probabilistic neural networks and non-linear transformation of textural features

Abstract

Introduction

Section snippets

Background and design considerations

Data acquisition

Experimental results

Discussion

Acknowledgement

Update Cancer Therapeut.

Surgery (Oxford)

Semin. Oncol. Nurs.

Magn. Reson. Imaging

Magn. Reson. Imaging

Magn. Reson. Imaging

J. Magn. Reson.

Comput. Biol. Med.

Artif. Intell. Med.

J. Magn. Reson.

Magn. Reson. Imaging

Comp. Graph. Image Proc.

Neural Networks