YiruMu

[YiruMu @mu_yiru]

This week's #SMILEJournalClub is on "RadImageNet: An Open Radiologic Deep Learning Research Dataset for Effective Transfer Learning." Is the RadImageNet pre-trained model a new way to go for transfer learning in brain health-related research?

[YiruMu @mu_yiru]

How do we determine whether a 'dataset' is good to use?

source: https://t.co/BYmYnjKkbM
##SMILEJournalClub

[YiruMu @mu_yiru]

This week's journal club is on a deep learning system(DeepDR) for detecting diabetic retinopathy across the disease spectrum. The DeepDR has three sub-networks that perform different tasks(image quality evaluation, lesion analysis, and DR grading).

[YiruMu @mu_yiru]

The system has shown promising clinical results. There are 4 main types of lesions as shown in the paper. However, by limiting the types to only 4, isn't it excluding the possibility of 'others'? Of course, the labels depending on the dataset.

[YiruMu @mu_yiru]

In the paper, the researchers added a direct comparison to a 3D DNN but they found that the additional results didn't change the pattern of results found before this comparison. What do you think the reasons might be?

[YiruMu @mu_yiru]

This week's journal club is on different scaling of linear models and deep learning in UKBioBank brain images versus machine-learning datasets. #SMILEJournalClub

[YiruMu @mu_yiru]

The paper seeks to answer the question of whether emerging large-scale brain-imaging datasets contain nonlinear information to better predict common phenotypes that can be exploited by the currently available kernel and deep models.

[YiruMu @mu_yiru]

Do you think how the data was prepared, as well as the sMRI modality chosen, would introduce any potential bias when it comes to performance comparison between SML and DL?

[YiruMu @mu_yiru]

This week's journal club is on Deep learning encodes robust discriminative neuroimaging representations to outperform standard machine learning. The article looked into the performance comparison between standard machine learning and deep learning for brain imaging analysis.

[YiruMu @mu_yiru]

The paper focused on the representational learning of DL and how DL can exploit nonlinearities in the data to generate superior task-discriminative representations for characterizing the brain. #SMILEJournalClub

[YiruMu @mu_yiru]

What do you think is the significance of this large, multi-modality dataset? How's it relevant to your research?

[YiruMu @mu_yiru]

This week's journal club is on Multimodal population brain imaging in the UK Biobank prospective epidemiological study. #SMILEJournalClub

[YiruMu @mu_yiru]

What is unique about this study is that it aims to consistently acquire high-quality imaging data from 100,000 predominantly healthy participants for the next decades and addresses the problem of early disease prediction.

[YiruMu @mu_yiru]

How exactly does Models Genesis reduce annotation effort? Does the effort of pertaining such as model take into consideration as well?

[YiruMu @mu_yiru]

In this week's journal club, we saw how Models Genesis, pretraining a model from different perspectives, outperforms learning from scratch, existing pre-trained 3D models, reducing annotation efforts, and more. #SMILEJournalClub

[YiruMu @mu_yiru]

How do their train the weights in the recurrent neural network in the predictive coding part?

[YiruMu @mu_yiru]

Pretty interesting paper on modeling ventral and dorsal pathways of the brain using a single and parallel pathways approach with predictive coding.
https://t.co/JCTwpzIkI5 #SMILEJournalClub

[YiruMu @mu_yiru]

Parallel pathways trained with CPC can model both ventral and dorsal areas and reflect the different functionalities of both areas. Additionally, contrastive predictive coding is added at the end of the CNN output in the architecture.

[YiruMu @mu_yiru]

There are some similarities among the results and seemingly poorer results on OASIS. What might be the causes? I wonder if the architectures mentioned in the paper can potentially be used on the current project.

[YiruMu @mu_yiru]

https://t.co/LD2lCMRvom
This week's journal club is on Convolutional neural networks for classification of Alzheimer's disease: Overview and reproducible evaluation #SMILEJournalClub

[YiruMu @mu_yiru]

This paper is really helpful in learning the main types of approaches(2d slice-level, 3d patch-level, ROI-based, subject-level) of CNN classification in AD. It is interesting to see the results using these approaches on the test sets.