Anatomical Priors for Image Segmentation via Post-processing with Denoising Autoencoders

Larrazabal, Agostina J.; Martínez, César; Ferrante, Enzo

doi:10.1007/978-3-030-32226-7_65

Cited by 39 publications

(24 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Given several stacked U-Nets, these authors proposed to add shortcut connections for each U-Net pair, thus generating the coupled U-Net. Similar ideas can easily be considered as extensions to our neural model, and additional post-processing steps (e.g., related to the use of Conditional Random Fields (Krähenbühl & Koltun, 2011) or denoising auto-encoders (Larrazabal, Martinez, & Ferrante, 2019) to filter noise in the generated segmentation maps) could also be employed in order to further improve results.…”

Section: Discussionmentioning

confidence: 99%

Semantic segmentation and colorization of grayscale aerial imagery with W‐Net models

et al. 2020

View full text Add to dashboard Cite

The semantic segmentation of remotely sensed aerial imagery is nowadays an extensively explored task, concerned with determining, for each pixel in an input image, the most likely class label from a finite set of possible labels. Most previous work in the area has addressed the analysis of high-resolution modern images, although the semantic segmentation of historical grayscale aerial photos can also have important applications. Examples include supporting the development of historical road maps, or the development of dasymetric disaggregation approaches leveraging historical building footprints. Following recent work in the area related to the use of fully-convolutional neural networks for semantic segmentation, and specifically envisioning the segmentation of grayscale aerial imagery, we evaluated the performance of an adapted version of the W-Net architecture, which has achieved very good results on other types of image segmentation tasks. Our W-Net model is trained to simultaneously segment images and reconstruct, or predict, the colour of the input images from intermediate representations. Through experiments with distinct data sets frequently used in previous studies, we show that the proposed W-Net architecture is quite effective in colouring and segmenting the input images. The proposed approach outperforms a baseline corresponding to the U-Net model for the segmentation of both coloured and grayscale imagery, and it also outperforms some of the other recently proposed approaches when considering coloured imagery. K E Y W O R D S fully-convolutional neural networks, processing grayscale aerial photos, semantic segmentation of remotely sensed imagery, W-Net architecture 1 | INTRODUCTION Large amounts of high-resolution remote sensing images are nowadays acquired daily through satellites and aerial vehicles, and used as base data for mapping and Earth observation activities. An intermediate step for converting these raw images into map layers in vector format is semantic image segmentation, which is nowadays an extensively explored task concerned with determining, for each pixel, the most likely class label from a finite set of possible labels, corresponding to the desired object categories to map (e.g., discriminating pixels referring to roads, buildings, or vegetation, considering inputs such as high-resolution images depicting urban areas). Despite the interest in the area and the many recent significant advancements, the task remains quite challenging for automated approaches. In the particular case of urban areas, semantic segmentation involves not only dealing with objects that may be partially obscured by cloud coverage, but also with the fact that objects in cities can be small,

show abstract

Section: Discussionmentioning

confidence: 99%

Semantic segmentation and colorization of grayscale aerial imagery with W‐Net models

et al. 2020

View full text Add to dashboard Cite

show abstract

“…A group of semi-supervised methods has addressed semi-supervised segmentation by incorporating prior knowledge/ domain knowledge such as anatomical priors about the objects of interest into the segmentation model as a strong regularization [91]- [96]. In fact, prior knowledge about location, shape, anatomy, and context is also crucial for manual annotation, especially in the presence of fuzzy boundaries or low image contrast.…”

Section: Prior Knowledge Learningmentioning

confidence: 99%

Medical Image Segmentation With Limited Supervision: A Review of Deep Network Models

Wang

2021

IEEE Access

View full text Add to dashboard Cite

Despite the remarkable performance of deep learning methods on various tasks, most cutting-edge models rely heavily on large-scale annotated training examples, which are often unavailable for clinical and health care tasks. The labeling costs for medical images are very high, especially in medical image segmentation, which typically requires intensive pixel/voxel-wise labeling. Therefore, the strong capability of learning and generalizing from limited supervision, including a limited amount of annotations, sparse annotations, and inaccurate annotations, is crucial for the successful application of deep learning models in medical image segmentation. However, due to its intrinsic difficulty, segmentation with limited supervision is challenging and specific model design and/or learning strategies are needed. In this paper, we provide a systematic and up-to-date review of the solutions above, with summaries and comments about the methodologies. We also highlight several problems in this field, discussed future directions observing further investigations.

show abstract

“…Figure 5 depicts the layers of network employed to train the segmentation model. Compared to that of CNN approach [15,16], this approach is significantly more abridged and upfront. This is due to the approach does not employ convolutional layers in CNN as it can make use of the amplified individual pixel information to create an in-depth relationship in the neural networks employed.…”

Section: 3deep Learning Neural Networkmentioning

confidence: 99%

Amplification of pixels in medical image data for segmentation via deep learning object-oriented approach

Fadzil¹,

Khalid²,

Ibrahim³

2021

IJATEE

View full text Add to dashboard Cite

Medical images serve as a very important tool for medical diagnosis. Medical image segmentation is an area of image processing that segments critical parts of a medical image for diagnosis purposes. The emergence of machine learning approach for medical image segmentation specifically by employing Convolutional Neural Network (CNN) has become a ubiquity as other approaches does not able to compete with its robustness and accuracy. However, this approach is very exhaustive in terms of time and computing resources. The CNN approach mostly emphasizes on the spatial information regarding the image without using much of the individual data contained withing the image. Therefore, this paper proposed a method to amplify the pixel data of medical images via Object-oriented Programming (OOP) approach for segmentation using a straightforward sequential deep learning approach. The results indicated that the proposed method allows more than 90 % faster training time with 33.8 seconds average and overall better segmentation performance of 0.744 for balanced-accuracy metric compared to recent state-of-the-art CNN segmentation models such as SegNet and U-Net Models.

show abstract

Anatomical Priors for Image Segmentation via Post-processing with Denoising Autoencoders

Cited by 39 publications

References 10 publications

Semantic segmentation and colorization of grayscale aerial imagery with W‐Net models

Semantic segmentation and colorization of grayscale aerial imagery with W‐Net models

Medical Image Segmentation With Limited Supervision: A Review of Deep Network Models

Amplification of pixels in medical image data for segmentation via deep learning object-oriented approach

Contact Info

Product

Resources

About