LEE- Biyomedikal Mühendisliği-Yüksek Lisans
Bu koleksiyon için kalıcı URI
Gözat
Son Başvurular
1 - 5 / 14
-
ÖgeKatmanlı ortamlarda fotoakustik görüntüleme için ses hızı tahmin yöntemi geliştirilmesi(Lisansüstü Eğitim Enstitüsü, 2025-01-13)Mühendislik ve sağlık alanlarında önemli yeniliklere olanak sağlayan fotoakustik görüntüleme, optik görüntülemenin yüksek kontrast ve spektroskopik özgüllüğünü, ultrason görüntülemenin uzamsal çözünürlüğüyle bir araya getiren hibrit bir modalitedir. Laser (light amplification by stimulated emission of radiation) ışığı ile dokuya enerji verilmesi sonucunda oluşan termal genleşmenin akustik dalgalar üretmesi prensibine dayanır. Bu akustik dalgalar, dedektörler tarafından algılanır ve biyolojik dokuların optik ve akustik özelliklerini yüksek çözünürlükle haritalayan görüntüler oluşturulur. Oluşturulan görüntülerin niteliği özellikle biyomedikal alanında doğrudan sağlığı etkileyeceği için ayrı bir öneme sahiptir. Görüntüsü elde edilecek dokuya gönderilen laser atımının özellikleri ve oluşan dalgaların vücut içindeki hızlarının doğru bir şekilde ölçülmesi görüntü niteliği için önemli verilerdir. Organların homojen olmaması, kas ve yağ yoğunluğunun organdan organa veya aynı organın farklı bireylerde eşit dağılmaması ve organların katmanlı yapıya sahip olması, akustik dalgaların hızlarının değişken olmasına neden olmaktadır, fakat gerçekte farklı hızlarla hareket eden bu dalgalar sanki homojen bir ortamda hareket ediyor gibi eşit hıza sahip olduğu varsayılmaktadır. Bu durum, görüntü niteliğinin bozulmasına neden olmaktadır. Bu durum, literatürde birçok ses hızı tahmin yönteminin geliştirilmesine neden olmuştur. Bu sayede, çeşitli tekniklerle tek bir katmanın hızını belirleyen veya birden fazla katmanın hızını hesaplayan yöntemler ortaya konmuştur. Bu çalışmada, katmanlı ortamlarda her bir katmanın ses hızını ayrı ayrı tahmin edebilen bir yöntem geliştirilmiştir. Bu yöntemde katmanlı bir ortam oluşturulmuş ve son katmana bir kaynak yerleştirilmiştir. Kaynaktan gelen fotoakustik sinyallerin dönüştürücü(transdüser) dizi elemanlarına gelen fotoakustik sinyallerin ulaşma süreleri ve dönüştürücü dizi elemanlarına ulaşıncaya kadar aldıkları yol bilgisi kullanılarak, yolun hız ve zamanın çarpımı formülünden yola çıkarak ortamın ses hızını tahmin eden yöntem geliştirildi, fakat katmanlı ortamlarda kaynaktan çıkan fotoakustik (FA) dalgaları katmanların ses hızlarına bağlı olarak kırınıma uğrayacağı için ses hızlarının yanı sıra FA dalgalarının her bir dönüştürücü elemanına gelirken alınan yol da bilinmemektedir. Bu nedenle problem doğrusal olmayan bir problem olmaktadır. Problemi doğrusal hale getirmek için alınan yolu modelleyen yeni bir dalga yayılım modeli geliştirildi. Kaynak çıkan dalgaların katmanlardan geçişleri Snell yasası ile modellendi. Snell yasasında kırınım acısı da ses hızına bağlı olduğu için tahmini bir başlangıç değeri ile başlandı. Bu değerler, katmanların ses hızlarının alabileceği maksimum ve minimum değerler içerisinde kalınarak yapıldı. Bu şekilde tahmini alınan yol bilgileri belirlendi. Burada kaynağın dönüştürücü dizisinin orta elemanın tam karşısında olduğu varsayımı yapılarak n elemanlı dizi için problem simetrik hale getirildi. m katmanlı yapı için nXm boyutlu matris haline getirilen problemin bilinmeyen ses hızlarının çözümü için matris sayısına bağlı olarak en küçük kareler yöntemiyle çözüldü. Yöntemin doğruluğunun ispatlanması için benzetim çalışmaları ve deneysel çalışma yapılmıştır. Benzetim çalışmaların gerçekleşmesi için bir Matlab aracı olarak kullanılan k-Wave ile yapılmış ve hız tahminlerinin doğruluğu, farklı katman kalınlıkları ile farklı hızlar denenerek test edilmiştir. Ayrıca, deneysel doğrulama amacıyla agar fantom maddesi kullanılarak laboratuvar ortamında deney yapılmıştır. Bu çalışma, biyomedikal görüntüleme ve deri gibi katmanlı dokuların teşhis süreçlerinde kullanılabilecek gelişmiş bir yaklaşım sunmayı hedeflemektedir.
-
ÖgeSuperpixel assisted deep neural network for breast tumor segmentation in ultrasound images(Graduate School, 2022)Breast cancer is the leading type of cancer diagnosed in women, according to data from the World Health Organization (WHO) in 2020. It is also the type of cancer that causes the most deaths in women, with around 685,000 deaths. In the diagnosis of breast cancer, ultrasound imaging has been used frequently. Tumor segmentation from breast ultrasound (BUS) images is significant for the success of subsequent analyses such as tumor detection, classification, and treatment planning in computer-aided diagnostic (CAD) systems. However, traditional segmentation approaches are challenging to apply automatically since tumor size, shape, and echo intensity vary significantly among BUS images. Deep learning-based segmentation approaches have great potential to reduce the workload of radiologists and operator dependence by automating tedious tasks. However, tumor segmentation remains a difficult task for these approaches because of the high speckle noise, artifacts, poor contrast, and intensity inhomogeneity in BUS images. Even though tumor segmentation in BUS images is challenging owing to the nature of ultrasound images, accurate lesion segmentation by reducing human involvement is essential for easier breast cancer analysis and diagnosis. For this purpose, a new superpixel-assisted deep learning model is proposed, focusing on automatic binary class breast tumor segmentation. In this thesis, a superpixel-guided deep learning network, in which residual and channel attention blocks are integrated into the U-Net network, was proposed to address the above problems. The network contains a secondary input consisting of the corresponding superpixel images in addition to the main input comprising of BUS images. Firstly, the input image is over-segmented into primitive superpixel regions with texture consistency while less semantics using the simple linear iterative clustering (SLIC) algorithm to avoid speckle noise interference while enhancing the salience of tumors in the input image. Obtaining superpixels as a pre-segmentation method and feeding it as a second input to the network will provide good guidance in tumor segmentation. On the other hand, because there are far fewer superpixels than pixels in the input image, using superpixels can greatly lower the overall computational cost. Experiments in this thesis showed that the use of superpixel images can improve the tumor segmentation success of the proposed model. Images that were utilized to train and evaluate the breast tumor segmentation algorithm developed by this thesis were taken from two publicly available BUS image datasets, BUSI and UDIAT. A number of data augmentation and normalization techniques are applied to these datasets. In proposed U-Net-based model, residual blocks were placed in both encoder path and decoder path. Residual modules improve feature extraction and expression, as well as resolve degradation issues, allowing for greater accuracy gains with higher depth. Additionally, the output of residual blocks in the encoder part is passed through the Channel Attention (CA) block to increase the network's representational power. The channel attention mechanism can explore the interdependence between the feature channels to boost segmentation performance. The Superpixel Channel Attention (SCA) module is a combination of superpixel features and weighted channel information derived by the CA block. This module is a channel attention enriched module for integrating prior knowledge of superpixel. Furthermore, bottlenecks in U-shaped convolutional neural networks are a way to force the model to learn a compression of the input data. The idea is that this compressed view should only contain the useful information to be able to construct an output mask. Therefore, because the high-level feature map represents complex features with wide receptive fields and more channels, the Channel Attention Residual (CAR) module was added to the model's bottom layer. The training was repeated with the 5-fold cross-validation technique to obtain a more consistent model for all cases. The final pixel labels are voted in the final assessment using ensemble models with random parameter initializations in 5-fold data. It turns out that a final segmentation output from an ensemble of models trained with different inputs and using the same architecture outperforms a single model. Ensemble learning application on superpixel-guided deep learning network, which is the recommended approach for breast lesion segmentation, gives better results than all competing U-Net variant models and either of the 5-fold models. Test results within the scope of the thesis showed that tumor segmentation from breast ultrasound images can be effectively accomplished using a method that combines deep neural networks and superpixel information.
-
ÖgeSelf-supervised deep convolutional neural network training for low-dose CT reconstruction(Graduate School, 2022)Computed tomography (CT) is a medical imaging technique to obtain a noninvasively three-dimensional image of the body. CT imaging is one of the most crucial tools which is used for monitoring the human body and diagnosing serious illnesses. In CT imaging, one of the most serious concerns has been ionizing radiation since exposure to large amounts of radiation can cause serious illnesses. Various low-dose CT reconstruction approaches have been proposed to reduce the dose level without compromising image quality. With the emergence of deep learning, the increasing availability of computational power, and huge datasets, data-driven methods have recently gotten a lot of attention. Deep learning-based methods have also been applied in various ways to address low-dose CT reconstruction problem. However, the success of these methods is usually dependent on labeled data, which requires tedious work by radiologists for CT imaging. Recent studies, however, have also shown that training may be done successfully with noisy datasets without the requirement of noise-free target data. In this study, a training scheme is defined to use low-dose projections as their own training targets. We apply the self-supervision principle in the projection domain where the noise is element-wise independent, which is a requisite for self-supervised training methods. The parameters of a denoiser neural network are optimized through self-supervised training. Experiments are done with both analytical and human CT data. Slices from deep lesion dataset for human CT data and ellipses dataset for synthetic data are used. To simulate low-dose settings, 64 views parallel beam geometry is used. The noisy projections are created with additive white Gaussian noise with 30, 33, and 37 dB SNR values. The proposed method is compared with FBP, SART, SART+TV, SART+BM3D, and the supervised FBP+U-Net method. The methods are compared quantitatively with PSNR and SSIM metrics, and the reconstructions are qualitatively assessed regarding background smoothness, the sharpness of the details, and the recoverability of the lesions with some visual examples. In the comparisons, it is shown that the proposed method outperforms both traditional and compressed sensing-based iterative reconstruction methods in the reconstruction of analytic CT phantoms and real-world CT images in the low-dose CT reconstruction task, both qualitatively and quantitatively. Besides, it produces comparable results with the supervised approach.
-
ÖgeSkin lesion classification with machine learning(Graduate School, 2023)Skin lesions are the part of the skin that has an abnormal structure and appearance compared to the surrounding areas. While some skin lesions cause only a physically bad appearance, some may have cancer features. Nowadays, skin cancer is one of the most extensive cancers among humans. Therefore, the correct detection of cancerous lesions is of great importance in the treatment of skin cancer. Skin cancers are basically divided into two main types, melanoma which is dangerous, and non-melanoma. The low contrast between the lesioned and non-lesion areas in the images of melanoma skin cancers requires expertise in the application of diagnostic methods and involves relativity among dermatologists. In addition, it is very difficult to automatically analyze human skin due to geographical and climatic effects, roughness, tonal diversity, hair and many other complex structures, and low contrast problem. However, a successful automatic analysis system to be created will help dermatologists to diagnose and speed up the process. Technology advancements have made it possible for doctors to diagnose skin cancer from dermatoscopic images using computer-aided diagnosis techniques, such as deep learning and machine learning models. In this study, image processing and machine learning techniques were used to classify skin lesions. In image processing, different mathematical algorithms have been applied to increase image quality. After an image preprocessing stage, which includes filtering the unwanted pixels in the images, image segmentation was performed using the watershed method and the lesioned regions were separated. Then, based on the ABCDT rule, feature extraction was performed with the lesions, asymmetry, border irregularity, color, diameter, and texture analysis. Texture analysis was performed based on Haralick texture properties. Finally, classification was performed with softmax regression, k-nearest neighbor (KNN), and support vector machines (SVM) algorithm. Standard performance measures called accuracy, precision, recall, and F1-score values are used to evaluate the results of the methods used for classification. When the results obtained from the classifiers are compared, it has been observed that the accuracy of the SVM classifier is higher than the softmax regression and k-nearest neighbor (KNN) algorithm. Keywords: Skin lesion, skin cancer, image processing, ABCDT analysis, machine learning, support vector machine algorithm, softmax regression, k-nearest neighbor algorithm.
-
Öge1H-MRSI of the deep gray matter structures in patients with amyotrophic lateral sclerosis(Graduate School, 2024-08-19)Amyotrophic Lateral Sclerosis (ALS) is a life-threatening disease that causes degeneration in nerve cells in the brain and spinal cord. In ALS patients, voluntary control of the arms and legs are affected. Currently, there is no cure for ALS. The primary goal of treatment is to manage symptoms to the greatest extent possible. Magnetic Resonance Spectroscopy (MRS) is employed to assess the concentration of metabolites in specific brain regions. This method has not been fully explored to understand the metabolic deficits in individuals diagnosed with ALS. It is crucial to understand the metabolic effects of ALS through various brain regions, particularly at the deep gray matter structures. 1H-MRSI data for 30 ALS patients, with a mean age of 57.8±9.55 years (17 females, 13 males) and 27 healthy controls, with a mean age of 48.44±10.5 years (16 females, 11 males) were acquired using a clinical 3T Siemens magnetic resonance imaging (MRI) scanner equipped with a multivoxel semi-LASER (sLASER) sequence (TR=1700ms, TE=40ms, VOI=10x10x15 mm3). In this study it is aimed to assess the metabolic differences between ALS patients and healthy controls (HC) at the thalamus, putamen, caudate and white matter regions. To achieve this, multivoxel magnetic resonance spectral data from ALS patients and HC were analyzed and metabolites Glx, GSH, tNAA, tCho, and Tau, as well as their ratio to tCr were quantified using LCModel. A Cramer-Rao lower bound (CRLB) threshold of less than 30 was employed to identify properly fitted metabolites. The metabolite peak ratios at the thalamus, putamen, caudate, and white matter regions were then compared between these two groups. A Wilcoxon signed-rank test was used to detect statistically significant differences in metabolite peak ratios between the left and right hemispheres at the thalamus, putamen, caudate, and white matter regions. The Mann-Whitney rank-sum test was used to evaluate metabolite peak ratio differences between ALS patients and healthy controls at some deep gray matter structures and white matter. MNI152 brain atlas was consulted to define the thalamus, putamen, and caudate regions. In conclusion, in this study, it was observed that ALS patients had a higher tCho/tCr ratio at the putamen compared to HC. Additionally, there was a trend for a lower Glx/tCr ratio at the left putamen of ALS patients compared to HC. ALS patients also showed a trend towards higher tCho/tCr and GSH/tCr ratios at the left caudate compared to HC. In ALS patients, higher Glx/tCr ratios were observed in the right thalamus and putamen compared to the left side. A trend towards a lower Glx/tCr ratio at the right caudate of ALS patients was observed compared to the left side. Additionally, ALS patients showed a lower tNAA/tCr ratio at the right caudate compared to the left side. In HC, higher Glx/tCr ratios were observed at the right thalamus compared to the left side. A trend towards higher tNAA/tCr and Glx/tCr ratios at the right putamen and right white of HC compared to the left sides were also observed. In this study, metabolic alterations were detected at the deep gray matter regions associated with executive function and behavior. Results of this study showed increased gliosis because of increased tCho/tCr ratio in ALS patients, a response to oxidative stress because of elevated GSH/tCr ratio in ALS patients, and deficiency in glutamate within these structures in individuals with ALS. The results of this study can contribute to a deeper understanding of ALS. Moreover, the findings obtained from this study suggest that MRS is a significant diagnostic and monitoring tool for neurodegenerative diseases such as ALS.