Literature Survey (1)
Literature Survey (1)
Title: MRI Brain Tumor Detection Methods Using Contourlet Transform Based on Time Adaptive Self-
Organizing Map
Authors: Ali Farzamnia, Seyed Hamidreza Hazaveh, Seyede Safieh Siadat, Ervin Gubin Moung
Explanation:
This paper introduces a novel method for detecting and classifying brain tumors in MRI images using
a combination of Contourlet Transform and a Time Adaptive Self-Organizing Map (TASOM). The
classification process is further optimized using the Whale Optimization Algorithm (WOA). The
Contourlet Transform is effective in capturing fine directional details and edge-based texture features
in MRI images, which are crucial for differentiating between benign and malignant tumors. The
TASOM improves adaptability and learning speed during classification. The proposed method
achieves an impressive classification accuracy of over 98.5%, with a strong balance between
precision and execution time. Comparative studies against traditional techniques indicate that this
hybrid model enhances both diagnostic performance and computational efficiency, making it a
promising tool for supporting radiologists in clinical settings.
---
Paper 2
Title: Brain Tumor and Glioma Grade Classification Using Gaussian Convolutional Neural Network
Authors: Muhammad Rizwan, Aysha Shabbir, Abdul Rehman Javed, Maryam Shabbir, Thar Baker,
Dhiya Al-Jumeily OBE
Explanation:
This paper presents an approach based on Gaussian Convolutional Neural Network (GCNN) for brain
tumor classification and glioma grade prediction. The study uses two MRI datasets: one for
classifying tumors into pituitary, glioma, and meningioma, and the second for classifying glioma into
grades II, III, and IV. The datasets contain T1-weighted contrast-enhanced images and include a total
of 3064 and 516 images respectively. The GCNN is designed to enhance image features using
Gaussian-based filters before applying convolution, which helps in better spatial localization of
tumors. The proposed model achieves an accuracy of 99.8% on the tumor type classification task and
97.14% on glioma grading. These results highlight the capability of GCNN to perform multi-class
classification efficiently. The model can serve as a reliable tool for early diagnosis and tumor grading,
which is crucial for treatment planning.
---
Paper 3
Title: A Deep Learning Model Based on Concatenation Approach for the Diagnosis of Brain Tumor
Authors: Neelum Noreen, Sellappan Palaniappan, Abdul Qayyum, Iftikhar Ahmad, Muhammad
Imran, Muhammad Shoaib
Explanation:
This paper proposes a deep learning approach that uses multi-level feature extraction and
concatenation for accurate brain tumor classification from MRI images. The method leverages two
powerful pre-trained convolutional neural networks: Inception-v3 and DenseNet201. Features are
extracted from different layers of each model to capture both low-level and high-level information.
These features are then concatenated and passed to a softmax classifier for final prediction. The
model is tested on a publicly available three-class brain tumor dataset, which includes glioma,
meningioma, and pituitary tumors. The proposed approach achieves 99.34% accuracy using
Inception-v3 and 99.51% using DenseNet201. The fusion of diverse feature representations enhances
the model’s ability to distinguish between tumor types. The study proves that combining multiple
deep learning models can significantly outperform single-model approaches, offering a robust
framework for automated brain tumor diagnosis.
Paper 4
Title: Improved Brain Tumor Segmentation and Classification in Brain MRI with FCM-SVM: A
Diagnostic Approach
Authors: Samar M. Alqhtani, Toufique Ahmed Soomro, Ahmed Ali Shah, Abdul Aziz Memon,
Muhammad Irfan, Saifur Rahman, Mohammed Jalahah, Abdulkarem H. M. Almawgani, and Ladon
Ahmed Bade Eljak
Explanation:
This study introduces an automated diagnostic system that significantly enhances the segmentation
and classification of brain tumors in MRI images using a hybrid method based on Fuzzy C-Means
(FCM) for segmentation and Support Vector Machine (SVM) for classification. The approach begins
with preprocessing steps that include Contrast Limited Adaptive Histogram Equalization (CLAHE) and
diffusion filtering to enhance the visual quality of MRI scans. Then, FCM clustering identifies the
tumor region with high spatial accuracy. The segmented region is passed to the SVM classifier, which
differentiates among tumor types such as meningioma, glioma, and pituitary tumors. The model uses
the CE-MRI dataset and delivers highly accurate results with accuracy of 98.2%, sensitivity of 97.7%,
specificity of 97.9%, and a Dice Score (DSC) of 96.1%, all while maintaining a fast processing time of
just 0.42 seconds per image. Compared to other state-of-the-art models, this method demonstrates
superior performance and provides a solid foundation for clinical decision-making in brain tumor
diagnosis.
---
Paper 5
Explanation:
This paper proposes SSBTCNet, a novel semi-supervised deep learning model for classifying brain
tumors in MRI images into three categories — glioma, meningioma, and pituitary tumors — and also
identifying non-tumor cases. The architecture integrates an AutoEncoder (AE) and a Multi-Layer
Perceptron (MLP) classifier trained simultaneously to learn meaningful representations. Unlike
conventional supervised models that rely solely on labeled data, SSBTCNet effectively utilizes
unlabeled data as well, improving generalization and learning efficiency. A fuzzy-logic-based
enhancement technique generates augmented training samples, making the system robust against
variations in image orientation, scale, and flip. This dual strategy (semi-supervised learning and data
augmentation) significantly boosts model performance. Evaluations on multiple datasets, with and
without data augmentation, reveal that SSBTCNet outperforms many traditional and deep learning-
based classification methods. Its ability to work efficiently even with limited labeled data makes it
ideal for practical scenarios where data annotation is scarce, enhancing early and accurate brain
tumor detection.
---
Paper 6
Title: Abnormal Brain Tumors Classification Using ResNet50 and Its Comprehensive Evaluation
Authors: Ayesha Younis, Qiang Li, Zargaam Afzal, Mohammed Jajere Adamu, Halima Bello Kawuwa,
Fida Hussain, Hamid Hussain
Explanation:
This study utilizes a deep learning approach with ResNet50, a robust convolutional neural network
architecture, to classify brain tumors from MRI images into glioma, meningioma, pituitary tumors,
and no tumor. The model was trained on 5712 MRI images and tested on 1311 images, incorporating
data augmentation and transfer learning techniques to improve performance and reduce overfitting.
The model achieved 99% accuracy on both training and validation datasets. Class-specific accuracies
were remarkable: 98.33% for glioma, 94.44% for meningioma, and 100% for both pituitary and no
tumor. The method also demonstrated strong performance across evaluation metrics, including a
precision of 0.9836, recall of 0.9832, F1 score of 0.9831, and ROC-AUC of 0.9990. These results
indicate excellent classification capability and robustness. The study emphasizes the effectiveness of
using transfer learning with ResNet50 for complex medical imaging tasks, highlighting its potential to
be integrated into clinical workflows for accurate and fast brain tumor diagnosis.
model demonstrates the benefits of ensemble learning in medical imaging, offering higher diagnostic
reliability, although it requires significant computational resources for training.
Posed
PAPER 7
Title: Brain Tumor Categorization and Retrieval Using Deep Brain Incep Res Architecture Based
Reinforcement Learning Network
Year: 2023
Explanation: This paper presents a novel Deep Brain Incep Res Architecture 2.0-based Reinforcement
Learning Network (DBIRA2.0-RLN) for categorizing and retrieving brain tumor images from MRI data.
The method integrates Reinforcement Learning (RL) with Deep Learning (DL) to classify three types
of brain tumors (meningioma, glioma, pituitary) and non-tumor images. The architecture utilizes an
Inception block with skip connections and incorporates a fuzzy logic-based technique for sample
improvement. The system achieved remarkable accuracy rates, including 100% for no tumor, 98.7%
for glioma, and 97.1% for meningioma, showcasing the efficiency and high performance of the
proposed approach compared to other methods in the field.
---
PAPER 8
Title: Optimized Brain Tumor Detection: A Dual-Module Approach for MRI Image Enhancement and
Tumor Classification
Year: 2023
Explanation: This paper introduces a dual-module approach for brain tumor detection that improves
both image enhancement and tumor classification. The first module employs adaptive Wiener
filtering, independent component analysis (ICA), and neural networks to enhance MRI images by
reducing noise and improving contrast. The second module uses Support Vector Machines (SVM) to
classify tumor types. The method demonstrated outstanding results, with accuracy as high as 98.9%
and Dice Score of 98.1%, offering a fast processing time of 0.43 seconds. This technique outperforms
existing methods in both sensitivity and specificity, making it a potential tool for clinical applications.
---
PAPER 9
Title: Improving Effectiveness of Different Deep Transfer Learning-Based Models for Detecting Brain
Tumors From MR Images
Year: 2023
Explanation: This paper explores the use of transfer learning techniques for classifying brain tumors
from MRI scans. Four deep learning models—Xception, NasNet Large, DenseNet121, and
InceptionResNetV2—were employed to extract features, with training performed on two benchmark
MRI datasets. The results showed that the Xception model with the ADAM optimizer outperformed
others, achieving accuracy rates up to 99.67% for a larger dataset. This study demonstrates the
effectiveness of deep transfer learning in improving classification accuracy, although performance
slightly decreased on smaller datasets, indicating the impact of data size on model efficiency.
---
PAPER 10
Title: Ensemble Deep Learning Models for Enhanced Brain Tumor Classification by Leveraging
ResNet50 and EfficientNet-B7
Year: 2023
Explanation: In this paper, an ensemble deep learning model combining ResNet50 and EfficientNet-
B7 is proposed for enhanced brain tumor classification. The ensemble model leverages the strengths
of both architectures, achieving superior performance in categorizing MRI images of brain tumors.
The model was trained on over 22,000 high-resolution images, and results showed that the
ensemble outperformed individual models, with a validation accuracy of 99.68%.