Investigation of biomedical image segmentation with deep learning methods
Students Name: Pasternak Diana Mykhailivna
Qualification Level: magister
Speciality: Information Technology Design
Institute: Institute of Computer Science and Information Technologies
Mode of Study: full
Academic Year: 2023-2024 н.р.
Language of Defence: англійська
Abstract: Pasternak D. M., Farmaha I. V. (supervisor). Investigation of biomedical image segmentation with deep learning methods. Master’s thesis. – Lviv Polytechnic National University, Lviv, 2023. Extended abstract. Medical image segmentation aims to make anatomical or pathological structures changes in more clear in images; it often plays a key role in computer aided diagnosis and smart medicine due to the great improvement in diagnostic efficiency and accuracy. To help clinicians make accurate diagnosis, it is necessary to segment some crucial objects in medical images and extract features from segmented areas [1][2]. Covering the segmentation approaches of various types of images within single research is pretty challenging task as selection of deep learning methods highly depends on data characteristics. Therefore, we decided to focus on segmentation of brain tumor MRI volumes to demonstrate power of neural networks as a front-line diagnostic tool as there has been comparatively less research in this area. Moreover, brain MRI volumes are a particularly intriguing subject for investigation due to their three-dimensional representation, voxel-wise processing, and the necessity for preprocessing with image characteristics such as contrast, brightness, etc. Gliomas are the most common type of primary brain tumors. Although gliomas are relatively rare, they are among the deadliest types of cancer, with a survival rate of less than 2 years after diagnosis. As the success of many brain tumor treatments depends on early intervention, early detection is paramount. In this context, an automated segmentation method for brain tumor segmentation is necessary as an efficient and reliable method for brain tumor detection and quantification [3]. In this study, we propose a fully automatic method for brain tumor segmentation, which is developed using U-Net and GAN based deep convolutional networks. Our method was evaluated on Multimodal Brain Tumor Image Segmentation Challenge (BRATS 2023) datasets. More complicated training schemes and neural network architectures (attention mechanism and deformable convolutions) were investigated without significant performance gain at the cost of greatly increased training time, and on the contrary with overfitting observed. Overall, our approach using original architectures with slight modifications yielded good and balanced performance for each tumor subregion with utilization of pixel-wise metrics and Dice score evaluation. Study object – segmentation of MRI images for brain tumor detection and creation of segmentation masks with tumor region classification. Scope of research – approach for brain tumor segmentation with help of architectures of deep neural networks, specifically convolutional and generative adversarial networks. Goal of research: generating accurate delineation of brain tumor regions; improving the accuracy of image segmentation by developing deep learning architectures based on existing networks; studying the influence of various structural parameters of models on segmentation accuracy; finding optimal configurations for each model. Results of brain tumor segmentation models performance are presented based on Dice scores values, Hausdorff distance, and pixel metrics of sensitivity and specificity. Attention U-Net 3D proved to be the best solution, considering the nature of the investigated dataset (~89.5% accuracy). The generative adversarial network (Vox2Vox) demonstrated an accuracy of nearly 86.3% (Dice coefficient). Keywords: brain tumor, medical image segmentation, deep learning, BraTS, U-Net, Generative Adversarial Networks, attention mechanism. References. [1] Risheng Wang, Tao Lei, Ruixia Cui, Bingtao Zhang, Hongying Meng and Asoke K. Nandi. Medical Image Segmentation Using Deep Learning: A Survey, 2021. arXiv:2009.13120v3. [2] Zhihua Liu, Lei Tong, Zheheng Jiang, Long Chen, Feixiang Zhou, Qianni Zhang, Xiangrong Zhang, Yaochu Jin, Huiyu Zhou. Deep Learning Based Brain Tumor Segmentation: A Survey, 2021. arXiv:2007.09479v3. [3] David Bouget, Andre Pedersen, Asgeir Store Jakola, Vasileios Kavouridis, Kyrre Eeg Emblem, Roelant S. Eijgelaar. Preoperative brain tumor imaging: models and software for segmentation and standardized reporting, 2022. arXiv:2204.14199v1.