AAEGAN Loss Optimizations Supporting Data Augmentation on Cerebral Organoid Bright-field Images

Martin, Clara Brémond; Chane, Camille Simon; Clouchoux, Cédric; Histace, Aymeric

doi:10.5220/0010780000003124

Cited by 5 publications

(16 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Computer graphics primarily aims to generate visually realistic graphics using computers. To achieve this, it involves establishing a geometric representation of the scene depicted in the graphics [18]. Subsequently, lighting models are employed to calculate the lighting effects under hypothetical light sources, considering factors such as texture and material properties [3].…”

Section: Utilization Goal Of Computer Graphicsmentioning

confidence: 99%

Investigation into the Application of Image Modeling Technology in the Field of Computer Graphics

Doungmala¹

2023

Int. J. Appl. Inf. Manag.

View full text Add to dashboard Cite

The field of computer graphics has experienced remarkable advancements in hardware and software, leading to rapid progress and development. One crucial aspect that significantly influences the authenticity and immersive nature of simulation environments is the utilization of image modeling technology. The success of creating realistic and captivating image modeling environments relies heavily on the effective implementation of this technology. In light of this, the present paper thoroughly examines the fundamental concepts of computer graphics and image modeling technology. Furthermore, it delves into an in-depth analysis of the integration of image modeling technology within the domain of computer graphics, followed by an exploration of the concealed algorithm behind its functioning. The continuous evolution of computer graphics systems, encompassing advancements in both hardware and software, has propelled the field forward at a remarkable pace. Among the various elements that contribute to the credibility and engagement of simulation environments, the utilization of image modeling technology stands out as a critical factor. This research paper endeavors to comprehensively investigate the fundamental principles of computer graphics and the intricate workings of image modeling technology. By dissecting and analyzing the conceptual framework of computer graphics and image modeling, the study sheds light on their interplay and interdependence. Finally, the paper unveils the concealed algorithm that underlies the operation of image modeling technology within the realm of computer graphics, thereby providing valuable insights into its inner workings.

show abstract

Section: Utilization Goal Of Computer Graphicsmentioning

confidence: 99%

Investigation into the Application of Image Modeling Technology in the Field of Computer Graphics

Doungmala¹

2023

Int. J. Appl. Inf. Manag.

View full text Add to dashboard Cite

show abstract

“…Data augmentation solutions have already been used to increase the size and diversity of this brain organoid bright-field dataset. An adversarial autoencoder (AAE) seems the architecture the most suited to augment images of brain organoid bright-field acquisition (Brémond Martin et al, 2022a ). This AAE differs from the original GAN architecture by the input given to the encoding part (original images) and its generative network containing an auto-encoder-decoder framework (Goodfellow et al, 2014 ; Makhzani et al, 2016 ).…”

Section: Introductionmentioning

confidence: 99%

“…To improve the sharpness during the generation, we test various loss functions to improve the adversarial network (Brémond Martin et al, 2022a ). However, these results are based upon metric calculation and a dimensional reduction to compare all feature images (original and generated with each optimization) in the same statistical space.…”

Section: Introductionmentioning

confidence: 99%

“…However, these results are based upon metric calculation and a dimensional reduction to compare all feature images (original and generated with each optimization) in the same statistical space. Indeed some metrics may not been suited to identify the naturality of an image as they are originally created to test the similitude or the quality of images (Borji, 2019 ; Brémond Martin et al, 2022a ). In our previous contribution we observe the data augmentation strategy based upon AAE loss optimizations used during the training step of a DL segmentation algorithm improve the quality of the shape extraction of brain organoids (Brémond Martin et al, 2022a ).…”

Section: Introductionmentioning

confidence: 99%

“…Indeed some metrics may not been suited to identify the naturality of an image as they are originally created to test the similitude or the quality of images (Borji, 2019 ; Brémond Martin et al, 2022a ). In our previous contribution we observe the data augmentation strategy based upon AAE loss optimizations used during the training step of a DL segmentation algorithm improve the quality of the shape extraction of brain organoids (Brémond Martin et al, 2022a ). The first raised question is does these images seems as natural as the original images to furnish a better segmentation quality compared to a result issue from classic data augmentation strategies?…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Brain organoid data synthesis and evaluation

Brémond-Martin,

Simon-Chane,

Clouchoux

et al. 2023

Front. Neurosci.

Self Cite

View full text Add to dashboard Cite

IntroductionDatasets containing only few images are common in the biomedical field. This poses a global challenge for the development of robust deep-learning analysis tools, which require a large number of images. Generative Adversarial Networks (GANs) are an increasingly used solution to expand small datasets, specifically in the biomedical domain. However, the validation of synthetic images by metrics is still controversial and psychovisual evaluations are time consuming.MethodsWe augment a small brain organoid bright-field database of 40 images using several GAN optimizations. We compare these synthetic images to the original dataset using similitude metrcis and we perform an psychovisual evaluation of the 240 images generated. Eight biological experts labeled the full dataset (280 images) as syntetic or natural using a custom-built software. We calculate the error rate per loss optimization as well as the hesitation time. We then compare these results to those provided by the similarity metrics. We test the psychovalidated images in a training step of a segmentation task.Results and discussionGenerated images are considered as natural as the original dataset, with no increase of the hesitation time by experts. Experts are particularly misled by perceptual and Wasserstein loss optimization. These optimizations render the most qualitative and similar images according to metrics to the original dataset. We do not observe a strong correlation but links between some metrics and psychovisual decision according to the kind of generation. Particular Blur metric combinations could maybe replace the psychovisual evaluation. Segmentation task which use the most psychovalidated images are the most accurate.

show abstract