A systematic review of deep learning data augmentation in medical imaging: Recent advances and future research directions

Islam, Tauhidul; Hafiz, Md. Sadman; Jim, Jamin Rahman; Kabir, Md. Mohsin; Mridha, M.F.

doi:10.1016/j.health.2024.100340

Cited by 9 publications

(1 citation statement)

References 126 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Image augmentation pre-processing is carried out by changing the pixel value scale to a smaller range using the Softmax activation function which is useful during model training, rotating the image within a range of ±30 degrees, shifting the width and height of the image by 10%, enlarging 20% and reversing horizontally [18]. In Figure 2 there is a Flatten layer which functions to change the output from the previous layer which has several dimensions (such as 7x7x2048) into one dimension (such as 100,352 elements).…”

Section: Image Augmentationmentioning

confidence: 99%

Developing a classification system for brain tumors using the ResNet152V2 CNN model architecture

Rhomadhon,

Ningtias

2024

J. Soft Comput. Explor.

View full text Add to dashboard Cite

According to The American Cancer Society, in 2021 there were 24,530 cases of brain and nervous system tumors. The National Cancer Institute reports that there are approximately 4.4 new cases of brain tumors per 100,000 men and women per year. Brain tumors can be detected using magnetic resonance imaging (MRI), a scanning tool that uses a magnetic field and a computer to record brain images and is able to provide clear visualization of differences in soft tissue such as white matter and gray matter. However, this cannot be done optimally because it still relies on manual analysis, so it cannot classify brain tumor types on larger datasets with the potential for error and a low level of accuracy. To accurately determine the type of brain tumor, a better classification method is needed. The aim of this study is to determine the accuracy of brain tumor calcification using the deep learning model. In this study, the classification of brain tumor types was carried out using the ResNet152V2 convolutional neural network (CNN) model which has a depth of 152 layers. The dataset used in this study was 7,023 MRI images of brain tumors consisting of 1,645 meningiomas, 1,621 gliomas, 1,757 pituitary and 2,000 normal. Research results show an accuracy value of 94.44%, so it can be concluded that the ResNet152V2 model performs well in classifying brain tumor images and can be used as a medium for physicians to more accurately diagnose brain tumor patients more accurately.

show abstract

Section: Image Augmentationmentioning

confidence: 99%

Developing a classification system for brain tumors using the ResNet152V2 CNN model architecture

Rhomadhon,

Ningtias

2024

J. Soft Comput. Explor.

View full text Add to dashboard Cite

show abstract

Dopaminergic PET to SPECT domain adaptation: a cycle GAN translation approach

Lopes,

Jiao,

Xue

et al. 2024

Eur J Nucl Med Mol Imaging

View full text Add to dashboard Cite

Purpose Dopamine transporter imaging is routinely used in Parkinson’s disease (PD) and atypical parkinsonian syndromes (APS) diagnosis. While [11C]CFT PET is prevalent in Asia with a large APS database, Europe relies on [123I]FP-CIT SPECT with limited APS data. Our aim was to develop a deep learning-based method to convert [11C]CFT PET images to [123I]FP-CIT SPECT images, facilitating multicenter studies and overcoming data scarcity to promote Artificial Intelligence (AI) advancements. Methods A CycleGAN was trained on [11C]CFT PET (n = 602, 72%PD) and [123I]FP-CIT SPECT (n = 1152, 85%PD) images from PD and non-parkinsonian control (NC) subjects. The model generated synthetic SPECT images from a real PET test set (n = 67, 75%PD). Synthetic images were quantitatively and visually evaluated. Results Fréchet Inception Distance indicated higher similarity between synthetic and real SPECT than between synthetic SPECT and real PET. A deep learning classification model trained on synthetic SPECT achieved sensitivity of 97.2% and specificity of 90.0% on real SPECT images. Striatal specific binding ratios of synthetic SPECT were not significantly different from real SPECT. The striatal left-right differences and putamen binding ratio were significantly different only in the PD cohort. Real PET and real SPECT had higher contrast-to-noise ratio compared to synthetic SPECT. Visual grading analysis scores showed no significant differences between real and synthetic SPECT, although reduced diagnostic performance on synthetic images was observed. Conclusion CycleGAN generated synthetic SPECT images visually indistinguishable from real ones and retained disease-specific information, demonstrating the feasibility of translating [11C]CFT PET to [123I]FP-CIT SPECT. This cross-modality synthesis could enhance further AI classification accuracy, supporting the diagnosis of PD and APS.

show abstract

Classification of AO/OTA 31A/B femur fractures in X-ray images using YOLOv8 and advanced data augmentation techniques

Marullo,

Ulrich,

Antonaci

et al. 2024

Bone Reports

View full text Add to dashboard Cite

A systematic review of deep learning data augmentation in medical imaging: Recent advances and future research directions

Cited by 9 publications

References 126 publications

Developing a classification system for brain tumors using the ResNet152V2 CNN model architecture

Developing a classification system for brain tumors using the ResNet152V2 CNN model architecture

Dopaminergic PET to SPECT domain adaptation: a cycle GAN translation approach

Classification of AO/OTA 31A/B femur fractures in X-ray images using YOLOv8 and advanced data augmentation techniques

Contact Info

Product

Resources

About