🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Mehrer et al. provide important insight into how deep neural networks learn by showing how weight initialization directly affect higher-level data representations. This work supports ensemble learning for complementary information. nature.com/articles/s4146… #smilejournalclub

Wen et al. analyze machine learning approaches to classify Alzheimer's disease. Their analysis provides insight into the importance transparency in scientific progress. The authors propose Clinica to improve consistency in future works. doi.org/10.1016/j.medi… #SMILEjournalclub

Bakhtiari et al. model the specializations of the ventral and dorsal pathways of the visual cortex to improve deep learning methods. They accomplish this using a self-supervised predictive loss in a single ANN. biorxiv.org/content/10.110… #SMILEJournalClub

Zhou et al. has constructed Generic Autodidactic Models (Models Genesis) to learn 3D medical imaging problems via self-supervision. This approach allows the models to learn better than 2D approaches based on ImageNet pretraining. #SMILEjournalclub doi.org/10.1016/j.medi…

Miller et al. introduce UKBiobank, a large research database that consists of 100,000 subjects. The data includes imaging, genetic, and EHR reports. Miller et al. also present neuroimaging results on the first 5,000 subjects. #SMILEJournalClub nature.com/articles/nn.43…

Abrol et al. compare standard machine learning and deep learning algorithms for identifying age and gender in structural MRIs. This study is important in showing the advantages of deep learning in the future of AI studies. #SMILEJournalClub nature.com/articles/s4146…

Schulz et al. compared the linear vs. nonlinear models on MRI data by sample size. They found that all models improve comparably with sample size. In comparison, linear models fell behind on standard machine learning datasets. #SMILEJournalClub nature.com/articles/s4146…

Pan et al. introduces a joint Disease-image-Specific-Network and Feature-consistency Generative Adversarial Network. They improve diagnosis by generating increased multi-modal data and by focusing on disease-relevant regions. #SMILEJournalClub DOI: 10.1109/TPAMI.2021.3091214

He et al. propose meta-matching to translate models from large datasets to unseen phenotypes in small datasets. They study this using resting-state functional connectivity with kernel ridge regression (KRR) and a fully connected DNN. #SMILEJournalClub nature.com/articles/s4159…

Littlejohns et al. discussed the growth of UK Biobank's data repository. This paper covers UKBiobank's multi-modal imaging data for 100,000 participants. UKBiobank's imaging data represents the largest global study of its nature. #SMILEJournalClub nature.com/articles/s4146…

DOMINO: Domain-aware Model Calibration in Medical Image Segmentation deepai.org/publication/do… by Skylar E. Stolte et al. including Kyle Volle #Statistics #Estimator

I just arrived in Singapore for the #MICCAI2022 conference. I have the honor of representing SMILE Fang Lab with my oral talk on deep learning uncertainty in medical image segmentation.

I am so honored to have been able to present my work at #MICCAI2022. Thank you to MICCAI for a wonderful conference! Also big thanks to everything Ruogu Fang and UF BME have done to mentor me to reach my goals!

Dai et al. proposed DeepDR to detect screen for different levels of diabetic retinopathy. The method uses hard parameter sharing between subnetworks for DR screening, lesion detection and segmentation, and image quality. #SMILEJournalClub nature.com/articles/s4146…

Rombach et al. propose latent diffusion models (LDMs) for high-resolution image synthesis. Unlike pixel-wise models, LDMs operate in the latent space of pretrained autoencoders. This approach balances computational cost and performance. arxiv.org/pdf/2112.10752… #SMILEJournalClub

Qi et al. examined potential multimodal predictive biomarkers for Schizophrenia. Their study correlated polygenic risk scores in UKBiobank to smaller gray matter volume and decreased functional activation in frontotemporal cortex. nature.com/articles/s4146… #SMILEJournalClub

Mei et al. developed an alternative dataset to ImageNet for pretraining deep learning models using radiology images. RadImageNet performed better on average than ImageNet; hence, this work shows less dependence on weight initialization. doi.org/10.1148/ryai.2… #SMILEJournalClub

Groos et al. developed a deep learning method to predict unilateral and bilateral cerebral palsy (CP) using infants' spontaneous movements at 9-18 weeks. They assessed this with an international dataset of 557 infants' video recordings. #SMILEJournalClub ncbi.nlm.nih.gov/pmc/articles/P…

Athey and Wager estimated treatment using causal forests. This method was based on a cluster-robust random forests. The application was to predict the heterogeneity of learning interventions applied on National Study of Learning Mindsets. #SmileJournalClub arxiv.org/pdf/1902.07409…

I am very honored to have had my manuscript, "DOMINO: Domain-aware loss for deep learning calibration", accepted for publication in Software Impacts. DOMINO is a software that regularizes deep learning models to alleviate risk in high-impact applications. doi.org/10.1016/j.simp…