Attention-based deep learning segmentation: Application to brain tumor delineation

Karimzadeh, Reza; Fatemizadeh, Emad; Arabi, Hossein

doi:10.1109/icbme54433.2021.9750374

Cited by 30 publications

(4 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As we explore the field of brain tumor detection within the realm of advanced deep learning methodologies, it is essential to articulate the fundamental challenge our work aims to address. Brain tumor detection is a critical aspect of medical imaging, where traditional methods, often reliant on manual interpretation and conventional machine learning approaches, encounter obstacles related to scalability, subjectivity, and efficiency [10][11][12]. The overarching problem in this domain is the need for more accurate and efficient diagnostic tools to discern intricate patterns within complex brain anatomy.…”

Section: Literature Surveymentioning

confidence: 99%

Brain Tumor Detection for Efficient Adaptation and Superior Diagnostic Precision by Utilizing MBConv-Finetuned-B0 and Advanced Deep Learning

2024

IJIES

View full text Add to dashboard Cite

In the rapidly evolving landscape of medical imaging, our proposed work presents an innovative and efficient approach to brain tumor detection through advanced deep learning methodologies. Central to our methodology is the strategic utilization of pre-trained weights from the formidable MBConv-Finetuned-B0 model, initially honed on the expansive ImageNet dataset, providing a foundation rich in general visual knowledge. Our subsequent fine-tuning process targets specific layers relevant to brain tumor detection, introducing two distinct convolutional layers, MBConv 6, 55, and MBConv 6, 30, meticulously added to the MBConv-Finetuned-B0 base model. These layers are intricately designed to extract and refine features specific to brain tumors, ensuring a nuanced understanding of pathology and enhancing the model's discrimination and accuracy. The flexibility of our methodology is exemplified by the thoughtful consideration of two fine-tuning options: one that adjusts all layers of the model and another that selectively fine-tunes only the proposed layers. We conduct a detailed comparative analysis, including homogeneity and median feature values, placing our work in direct comparison with established techniques such as Ensemble Transfer Learning and Quantum Variational Classifier (ETL & QVC), Ultra-Light Deep Learning (ULDL) Model, Deep Convolutional Neural Network (DCNN), and Deep Learning and Image Processing (DLIP). The results showcase the model's proficiency, achieving an accuracy of 94%, precision of 84%, recall of 92%, F1 score of 88%, and an AUC-ROC of 96%. Notably, our model demonstrates superior performance in terms of homogeneity (vE Homogeneity: 0.93, vN Homogeneity: 0.91, Enhancement Homogeneity: 0.97) and median feature values (Median vE Feature Value: 0.82, Median vN Feature Value: 0.87, Median Enhancement Feature Value: 0.80), providing a comprehensive understanding of its effectiveness in capturing subtle nuances in brain tumor images.

show abstract

Section: Literature Surveymentioning

confidence: 99%

Brain Tumor Detection for Efficient Adaptation and Superior Diagnostic Precision by Utilizing MBConv-Finetuned-B0 and Advanced Deep Learning

2024

IJIES

View full text Add to dashboard Cite

show abstract

“…This combination of a U-Net architecture with attention mechanisms sets the DHA-ISSP model apart from previous methods that separately employed U-Net or attentionbased architectures. Karimzadeh et al [25] proposed AbUNet, utilizing the UNet architecture for efficient tumor segmentation. While AbUNet leverages the strengths of the UNet architecture, the DHA-ISSP model goes beyond by introducing dynamic hierarchical attention mechanisms and incorporating survival prognosis capabilities.…”

Section: Related Workmentioning

confidence: 99%

“…Myronenko [24] focused on boundary delineation but do not incorporate survival prediction capabilities. Karimzadeh et al's AbUNet [25] lacks the dynamic hierarchical attention mechanisms and survival prognosis capabilities of the DHA-ISSP model. The DHA-ISSP model stands out in the literature for its incorporation of dynamic hierarchical attention mechanisms, survival prognosis capabilities, and superior performance in brain tumor segmentation compared to existing methods.…”

Section: Related Workmentioning

confidence: 99%

“…VGG Net-Based DLF [16] Proposed Approach 3D U-net [18] Deep CNN and SVM [20] Ms 3D CNN [23] Attention-based DL [25] 0 While DenseNet and U-Net [29] are respected, they lack interpretability and clinical reliability. Our proposed model, designed for brain tumor segmentation and survival prediction, introduces Dynamic Hierarchical Attention (DHA) to capture crucial spatial information and highlight regions.…”

Section: Comparative Analysis Of Overall Performancementioning

confidence: 99%

See 1 more Smart Citation

Advancing Brain Tumor Analysis through Dynamic Hierarchical Attention for Improved Segmentation and Survival Prognosis

Kannan,

Anusuya

2023

CMC

View full text Add to dashboard Cite

Gliomas, the most prevalent primary brain tumors, require accurate segmentation for diagnosis and risk assessment. In this paper, we develop a novel deep learning-based method, the Dynamic Hierarchical Attention for Improved Segmentation and Survival Prognosis (DHA-ISSP) model. The DHA-ISSP model combines a three-band 3D convolutional neural network (CNN) U-Net architecture with dynamic hierarchical attention mechanisms, enabling precise tumor segmentation and survival prediction. The DHA-ISSP model captures fine-grained details and contextual information by leveraging attention mechanisms at multiple levels, enhancing segmentation accuracy. By achieving remarkable results, our approach surpasses 369 competing teams in the 2020 Multimodal Brain Tumor Segmentation Challenge. With a Dice similarity coefficient of 0.89 and a Hausdorff distance of 4.8 mm, the DHA-ISSP model demonstrates its effectiveness in accurately segmenting brain tumors. We also extract radio mic characteristics from the segmented tumor areas using the DHA-ISSP model. By applying cross-validation of decision trees to the selected features, we identify crucial predictors for glioma survival, enabling personalized treatment strategies. Utilizing the DHA-ISSP model and the desired features, we assess patients' overall survival and categorize survivors into short, mid, in addition to long survivors. The proposed work achieved impressive performance metrics, including the highest accuracy of 0.91, precision of 0.84, recall of 0.92, F1 score of 0.88, specificity of 0.94, sensitivity of 0.92, area under the curve (AUC) value of 0.96, and the lowest mean absolute error value of 0.09 and mean squared error value of 0.18. These results clearly demonstrate the superiority of the proposed system in accurately segmenting brain tumors and predicting survival outcomes, highlighting its significant merit and potential for clinical applications.

show abstract

Brain tumor detection and classification in MRI using hybrid ViT and GRU model with explainable AI in Southern Bangladesh

Ahmed,

Hossain,

Islam

et al. 2024

Sci Rep

View full text Add to dashboard Cite

Brain tumor, a leading cause of uncontrolled cell growth in the central nervous system, presents substantial challenges in medical diagnosis and treatment. Early and accurate detection is essential for effective intervention. This study aims to enhance the detection and classification of brain tumors in Magnetic Resonance Imaging (MRI) scans using an innovative framework combining Vision Transformer (ViT) and Gated Recurrent Unit (GRU) models. We utilized primary MRI data from Bangabandhu Sheikh Mujib Medical College Hospital (BSMMCH) in Faridpur, Bangladesh. Our hybrid ViT-GRU model extracts essential features via ViT and identifies relationships between these features using GRU, addressing class imbalance and outperforming existing diagnostic methods. We extensively processed the dataset, and then trained the model using various optimizers (SGD, Adam, AdamW) and evaluated through rigorous 10-fold cross-validation. Additionally, we incorporated Explainable Artificial Intelligence (XAI) techniques-Attention Map, SHAP, and LIME-to enhance the interpretability of the model’s predictions. For the primary dataset BrTMHD-2023, the ViT-GRU model achieved precision, recall, and F1-score metrics of 97%. The highest accuracies obtained with SGD, Adam, and AdamW optimizers were 81.66%, 96.56%, and 98.97%, respectively. Our model outperformed existing Transfer Learning models by 1.26%, as validated through comparative analysis and cross-validation. The proposed model also shows excellent performances with another Brain Tumor Kaggle Dataset outperforming the existing research done on the same dataset with 96.08% accuracy. The proposed ViT-GRU framework significantly improves the detection and classification of brain tumors in MRI scans. The integration of XAI techniques enhances the model’s transparency and reliability, fostering trust among clinicians and facilitating clinical application. Future work will expand the dataset and apply findings to real-time diagnostic devices, advancing the field.

show abstract

Attention-based deep learning segmentation: Application to brain tumor delineation

Cited by 30 publications

References 10 publications

Brain Tumor Detection for Efficient Adaptation and Superior Diagnostic Precision by Utilizing MBConv-Finetuned-B0 and Advanced Deep Learning

Brain Tumor Detection for Efficient Adaptation and Superior Diagnostic Precision by Utilizing MBConv-Finetuned-B0 and Advanced Deep Learning

Advancing Brain Tumor Analysis through Dynamic Hierarchical Attention for Improved Segmentation and Survival Prognosis

Brain tumor detection and classification in MRI using hybrid ViT and GRU model with explainable AI in Southern Bangladesh

Contact Info

Product

Resources

About