Browsing by Subject "Fully convolutional networks"
Now showing 1 - 3 of 3
Results Per Page
Sort Options
Item Open Access Deep convolutional network for tumor bud detection(Bilkent University, 2019-04) Koç, SonerThe existence of tumor buds is accepted as a promising biomarker for staging colorectal carcinomas. In the current practice of medicine, these tumor buds are detected by the manual examination of a immunohistochemically (IHC) stained tissue sample under a microscope. This manual examination is time-consuming as well as it may lead to inter-observer variability. In order to obtain fast and reproducible examinations, developing computational solutions has been becoming more and more important. With this motivation, this thesis presents a fully convolutional network design for the purpose of automatic tumor bud detection, for the rst time. This network design extends the U-net architecture by considering up-to-date learning mechanisms. These mechanisms include using residual connections in the encoder path, employing both ELU and ReLU activation functions in di erent layers of the network, training the network with a Tversky loss function, and combining outputs of di erent layers of the decoder path to reconstruct the nal segmentation map. Our experiments on 3295 image tiles taken from 23 whole slide images of IHC stained colorectal carcinomatous samples show that this extended version helps alleviate the vanishing gradient problem and those related with having a high class-imbalance dataset. And as a result, this network design yields better segmentation results compared with those of the two state-of-the-art networks.Item Open Access FourierNet: shape-preserving network for henle's fiber layer segmentation in optical coherence tomography images(Institute of Electrical and Electronics Engineers , 2023-02-06) Cansız, Selahattin; Kesim, C.; Bektaş, S.N.; Kulalı, Z.; Hasanreisoğlu, M.; Gündüz-Demir, ÇiğdemHenle's fiber layer (HFL), a retinal layer located in the outer retina between the outer nuclear and outer plexiform layers (ONL and OPL, respectively), is composed of uniformly linear photoreceptor axons and Müller cell processes. However, in the standard optical coherence tomography (OCT) imaging, this layer is usually included in the ONL since it is difficult to perceive HFL contours on OCT images. Due to its variable reflectivity under an imaging beam, delineating the HFL contours necessitates directional OCT, which requires additional imaging. This paper addresses this issue by introducing a shape-preserving network, FourierNet, which achieves HFL segmentation in standard OCT scans with the target performance obtained when directional OCT is available. FourierNet is a new cascaded network design that puts forward the idea of benefiting the shape prior of the HFL in the network training. This design proposes to represent the shape prior by extracting Fourier descriptors on the HFL contours and defining an additional regression task of learning these descriptors. FourierNet then formulates HFL segmentation as concurrent learning of regression and classification tasks, in which estimated Fourier descriptors are used together with the input image to construct the HFL segmentation map. Our experiments on 1470 images of 30 OCT scans of healthy-looking macula reveal that quantifying the HFL shape with Fourier descriptors and concurrently learning them with the main segmentation task leads to significantly better results. These findings indicate the effectiveness of designing a shape-preserving network to facilitate HFL segmentation without performing directional OCT imaging.Item Open Access Multi-task network for computed tomography segmentation through fractal dimension estimation(Bilkent University, 2023-01) Jabdaragh, Aziza SaberMulti-task learning proved to be an effective strategy to increase the performance of a dense prediction network on a segmentation task, by defining auxiliary tasks to reflect different aspects of the problem and concurrently learning them with the main task of segmentation. Up to now, previous studies defined their auxiliary tasks in the Euclidean space. However, for some segmentation tasks, the complexity and high variation in the texture of a region of interest may not follow the smoothness constraint in the Euclidean geometry. This thesis addresses this issue by introducing a new multi-task network, MTFD-Net, which utilizes the fractal geometry to quantify texture complexity through self-similar patterns in an image. To this end, we propose to transform an image into a map of fractal dimensions and define its learning as an auxiliary task, which will provide auxiliary supervision to the main segmentation task, towards betterment of left atrium segmentation in computed tomography images. To the best of our knowledge, this is the first proposal of a dense prediction network that employs the fractal geometry to define an auxiliary task and learns it in parallel to the segmentation task in a multi-task learning framework. Our experiments revealed that the proposed MTFD-Net model led to more accurate left atrium segmentation, compared to its counterparts.