Kyle See

Graduate Research Assistant at the University of Florida pursuing PhD in Biomedical Engineering

[Kyle See @kylebsee]

This paper, https://t.co/wgi1cB46sv, by Tang et al. introduces a self-supervised shifted window transformer for medical image analysis on a large computed tomography data set that is capable of extracting features at various resolutions from the whole image. #SMILEJournalClub

[Kyle See @kylebsee]

This paper, https://t.co/EAuNuXiMyr, by Athey and Wager explores an extension of random forests called causal forests and its' ability to understand how treatment effects vary across different samples. #SMILEJournalClub

[Kyle See @kylebsee]

This paper, https://t.co/eFGObx0OKt, by Groos et al. uses a deep learning method to predict cerebral palsy in infants from 9 to 18 weeks' corrected age. Their deep learning approach outperforms the clinically recommended general movement assessment (GMA) tool. #SMILEJournalClub

[Kyle See @kylebsee]

https://t.co/D2xtLDykOJ

[Kyle See @kylebsee]

A paper by Mei et al. introduces an open radiologic dataset called RadImageNet for use in transfer learning in medical applications. RadImageNet is only a tenth of the size of the widely used ImageNet and outperforms it in eight medical classification problems. #SMILEJournalClub

[Kyle See @kylebsee]

The authors introduce a joint polygenic risk score (PRS) and multimodal brain imaging study which classifies between schizophrenic (SZ) and healthy controls (HC). Their novel contribution is associating PRS-SZ with multimodal brain patterns to distinguish SZ and HC.

[Kyle See @kylebsee]

This week's #SMILEJournalClub focuses on a paper entitled, "Derivation and utility of schizophrenia polygenic risk associated multimodal MRI frontotemporal network" by Qi et al. https://t.co/Y8kxpaNLmD

[Kyle See @kylebsee]

...and multi-severity grading. This system can potentially serve as a cost-effective screening procedure that can significantly reduce the work load for ophthalmologists. https://t.co/Xm64Iy1M5G

[Kyle See @kylebsee]

This week's #SMILEJournalClub focuses on a paper by Dai et al. which developed a deep learning system to detect early-to-late stages of diabetic retinopathy. Their system, DeepDR, is trained with over 400k fundus images and achieves promising results for lesion detection...

[Kyle See @kylebsee]

This week's #SMILEJournalClub, "The UK Biobank imaging enhancement of 100,000 participants: rationale, data collection, management and future directions" by Littlejohns et al. revisits 100,000 participants with plans to expand the dataset with additional imaging modalities.

[Kyle See @kylebsee]

This paper, "Meta-matching as a simple framework to translate phenotypic predictive models from big to small data", by He et al. exploits phenotypic correlations to improve prediction performance of new phenotypes in small-scale studies. https://t.co/2iGWv36ENS #SMILEJournalClub

[Kyle See @kylebsee]

This week's journal club paper by Schulz et al. compares the sample size scaling of linear/non-linear machine learning and deep learning using the UK Biobank dataset and the MNIST and Fashion-MNIST benchmark datasets. https://t.co/diM8ne7ohr #SMILEJournalClub

[Kyle See @kylebsee]

An interesting paper by Abrol et al. emphasizes the importance of representation learning for deep learning in comparison to feature-selected standard machine learning approaches through multiple classification and regression tasks. https://t.co/mRpoQ7aeJb #SMILEJournalClub

[Kyle See @kylebsee]

The scope of this dataset includes various imaging modalities, health records, genetics, biological phenotyping, etc. The collection of this data has the potential to provide early insight into disease mechanisms and predictions.

[Kyle See @kylebsee]

This paper, "Multimodal population brain imaging in the UK Biobank prospective epidemiological study", by Miller et al. introduced a high quality imaging dataset of 100,000 healthy individuals. https://t.co/gUopHuq4hs #SMILEJournalClub

[Kyle See @kylebsee]

The authors model a deep artificial neural network to the ventral and dorsal pathways of the visual cortex. They employ a self-supervised predictive loss function and show that self-supervised learning has the potential to explain the functional properties of the visual cortex.

[Kyle See @kylebsee]

This week we look at a paper entitled, "The functional specialization of visual cortex emerges from training parallel pathways with self-supervised predictive learning", by Bakhtiari et al. https://t.co/R390kLqtR7 #SMILEJournalClub

[Kyle See @kylebsee]

In response, their contribution addressed the limitations across these studies. They published an open-source software called clinica. This software has multiple modular procedures for processing data from widely used datasets.

[Kyle See @kylebsee]

This paper, "Convolutional neural networks for classification of Alzheimer's disease: Overview and reproducible evaluation" by Wen et al. develops an open-source framework for classification of Alzheimer's disease using T1-MRIs. https://t.co/r9BLccAbqq #SMILEJournalClub

[Kyle See @kylebsee]

The authors reviewed studies that classify AD using T1-MRI and CNNs. They found that image preprocessing, feature selection, and other factors varied across studies. Although the studies provide insightful analysis of AD classification, many were not readily reproducible.