Defenseeeeeeeeedefenceeeeeunckefjdfd I.pptx

Defence of electrical and electronics

1. WELCOME TO PHASE-I

2. Supervised By- Dr. ASM Shihavuddin Chairman & Professor Department of EEE Green University of Bangladesh Course Code: EEE 400A Course Title: Project/Thesis Semester: Fall 2023 2

3. Presentation Team 3 Razaul Haque Rasham 201001049 Department of EEE Green University of Bangladesh Shimul Das 201001048 Department of EEE Green University of Bangladesh Md. Anwar Hossain 201001087 Department of EEE Green University of Bangladesh

4. Classification-Segmentation for 3D modeling and fusion technology of medical data via Transfer Learning and Residual Networks 2 4

5. TABLE OF CONTENTS 01 Objective of the research 02 Literature Review 03 Problem Definition 04 Scope of Work 05 Methodology 06 Gantt chart 07 Conclusion 08 Reference

6. Objective of The Research 6 T o explore and develop advanced techniques for processing and analyzing medical imaging data. Specifically, the thesis aims to investigate the use of transfer learning and residual networks for classification and segmentation tasks in 3D modeling and fusion of medical data. The primary goal is to improve the accuracy and efficiency of these tasks, which can be crucial in medical applications such as diagnosis, treatment planning, and surgical guidance. The thesis also aims to contribute to the growing field of machine learning in healthcare by exploring new methods for processing medical data and developing tools that can aid healthcare professionals in their work.

7. Literature Review 7 Author & Publication Date Title Methodology Used in Article Strength of the Research Limitation of the Research Mohsen Ahmadi, Abbas Sharifi, Mahta Jafarian Fard & Nastaran Soleimani. Published : 12 Feb 2021 Detection of brain lesion location in MRI images using convolutional neural network and robust PCA Deep learning method Convolutional Neural Network (CNN) Manual selection, No Database Sortation, seven different brain diseases classification Ramdas Vankdothu , Mohd Abdul Hameed Published : 23 August 2022 Brain tumor MRI images identification and classification based on the recurrent convolutional neural network [2] 3D-CNN, Prior algorithms 3D-CNN Manual selection, No Database Sortation Nghia Duong-Trung, Dung Ngoc Le Ha, Hiep Xuan Huynh Published : February 2022 Classifification -Segmentation Pipeline for MRI via Transfer Learning and Residual Networks [3] Transfer Learning, ResUnet Segmentation, Residual Networks Transfer Learning and Residual Networks Only use MRI Images, single clinical practice Author & Publication Date Title Methodology Used in Article Strength of the Research Limitation of the Research Mohsen Ahmadi, Abbas Sharifi, Mahta Jafarian Fard & Nastaran Soleimani. Published : 12 Feb 2021 Deep learning method Convolutional Neural Network (CNN) Manual selection, No Database Sortation, seven different brain diseases classification Ramdas Vankdothu , Mohd Abdul Hameed Published : 23 August 2022 Brain tumor MRI images identification and classification based on the recurrent convolutional neural network [2] 3D-CNN, Prior algorithms 3D-CNN Manual selection, No Database Sortation Nghia Duong-Trung, Dung Ngoc Le Ha, Hiep Xuan Huynh Published : February 2022 Classifification -Segmentation Pipeline for MRI via Transfer Learning and Residual Networks [3] Transfer Learning, ResUnet Segmentation, Residual Networks Transfer Learning and Residual Networks Only use MRI Images, single clinical practice

8. Problem Definition 8 Dataset bias: The performance of the model heavily relies on the quality and representativeness of the MRI dataset used for training. If the dataset is biased or does not fully capture the variability of the target population, the model may not generalize well to new data. Model overfitting: Deep learning models are prone to overfitting, which occurs when the model performs well on the training data but poorly on new data. This can be exacerbated if the dataset is small or if the model is too complex. Model interpretability: Deep learning models are often considered "black boxes" because it can be difficult to understand how the model arrives at its predictions. This lack of interpretability can make it challenging to identify and correct errors in the model's predictions. Computational requirements: Deep learning models, especially those based on transfer learning and residual networks, can be computationally intensive and require powerful hardware to train and deploy. This can limit the scalability and accessibility of the model. Ethical considerations: Medical imaging data is highly sensitive, and the use of deep learning models in healthcare raises important ethical considerations such as privacy, fairness, and bias. These issues must be carefully addressed to ensure that the model is deployed in a responsible and ethical manner.

9. Scope of Work 9 Dataset bias: Researchers should aim to collect a diverse and representative dataset that fully captures the variability of the target population. Additionally, they can use data augmentation techniques to increase the size and diversity of the dataset. Model overfitting: Researchers can use regularization techniques such as dropout or weight decay to prevent overfitting. They can also use early stopping or cross-validation to select the best model. Model interpretability: Researchers can use techniques such as activation maps or attention mechanisms to visualize and interpret the model's predictions. They can also use simpler models such as decision trees or logistic regression for better interpretability. Computational requirements: Researchers can use cloud-based services or distributed training to reduce the computational requirements of the model. They can also use model compression techniques such as pruning or quantization to reduce the model size and computational requirements. Ethical considerations: Researchers should carefully consider the ethical implications of their work and take steps to ensure that the model is deployed in a responsible and ethical manner. This can include obtaining informed consent, ensuring data privacy and security, and addressing potential sources of bias and discrimination.

10. 10 Methodology Step-8 Step-7 Step-6 Step-5 Step-4 Step-3 Step-2 Step-1 Literature Review Model Selection and Implementation Performance Evaluation Validation & Testing Data Collection and Processing Training and Testing the Models Optimization & Fine-Tuning Conclusion and Future Work

11. 12 Month Methodology 1 2 3 4 5 6 Literature Review ◆ Data Collection and Processing ◆ Model Selection and Implementation ◆ Training and Testing the Models ◆ Performance Evaluation ◆ Optimization & Fine-Tuning ◆ Validation & Testing ◆ Conclusion and Future Work ◆ Gantt chart 11

12. 12 Conclusion Improving the accuracy and efficiency of medical image analysis. By leveraging pre-trained models and residual connections, this pipeline is able to achieve superior results compared to traditional methods, and has the potential to enable more accurate diagnosis and treatment planning for a range of medical conditions. However, there are still challenges to be addressed, including the need for large and diverse datasets to fine-tune transfer learning models, and the need for more robust segmentation algorithms that can handle complex anatomical structures. With continued research and development, this pipeline has the potential to significantly improve the quality of care in medical imaging and diagnosis, and could ultimately help to save lives and improve patient outcomes.

13. 1 3 Resources [1] Mohsen Ahmadi, Abbas Sharifi, Mahta Jafarian Fard & Nastaran Soleimani, “ Detection of brain lesion location in MRI images using convolutional neural network and robust PCA” 12 Feb 2021, doi.org/10.1080/00207454.2021.1883602 [2] Ramdas Vankdothu , Mohd Abdul Hameed , “ Brain tumor MRI images identification and classification based on the recurrent convolutional neural network” 23 August 2022, doi.org/10.1016/j.measen.2022.100412 [3] Nghia Duong-Trung, Dung Ngoc Le Ha, Hiep Xuan Huynh, “ Classification-Segmentation Pipeline for MRI via Transfer Learning and Residual Networks” February 2022;DOI:10.15439/2021R14; Conference: International Conference on Research in Intelligent and Computing in EngineeringAt : Binh Duong, Vietnam [4] Rundo L, Han C, Zhang J, et al. CNN-based prostate zonal segmentation on T2-weighted MR images: a cross-dataset study. In Neural approaches to dynamics of signal exchanges. Springer; 2020. p. 269–280. [ Crossref ], [Google Scholar] [5] Hamzenejad A, Jafarzadeh Ghoushchi S, Baradaran V, et al. A robust algorithm for classification and diagnosis of brain disease using local linear approximation and generalized autoregressive conditional heteroscedasticity model. Mathematics. 2020;8(8):1268. [ Crossref ], [Web of Science ®], [Google Scholar] [6] Sun Y, Zhou C, Fu Y, et al. Parasitic GAN for semi-supervised brain tumor segmentation. In 2019 IEEE International Conference on Image Processing (ICIP), 2019. p. 1535–1539. [ Crossref ], [Google Scholar] [7] N. Duong-Trung, L.-D. Quach, M.-H. Nguyen, and C.-N. Nguyen, “ Acombination of transfer learning and deep learning for medicinal plantclassification ,” in Proceedings of the 2019 4th International Conferenceon Intelligent Information Technology, 2019, pp. 83–90.

14. Thank You