Published on Mon Nov 16 2020

The unreasonable effectiveness of Batch-Norm statistics in addressing catastrophic forgetting across medical institutions

Sharut Gupta, Praveer Singh, Ken Chang, Mehak Aggarwal, Nishanth Arun, Liangqiong Qu, Katharina Hoebel, Jay Patel, Mishka Gidwani, Ashwin Vaswani, Daniel L Rubin, Jayashree Kalpathy-Cramer

Model brittleness is a primary concern when deploying deep learning models in medical settings. Fine-tuning the model on subsequent institutions after training it on the original institution results in a decrease in performance. This phenomenon is called catastrophic forgetting.

0
0
0
Abstract

Model brittleness is a primary concern when deploying deep learning models in medical settings owing to inter-institution variations, like patient demographics and intra-institution variation, such as multiple scanner types. While simply training on the combined datasets is fraught with data privacy limitations, fine-tuning the model on subsequent institutions after training it on the original institution results in a decrease in performance on the original dataset, a phenomenon called catastrophic forgetting. In this paper, we investigate trade-off between model refinement and retention of previously learned knowledge and subsequently address catastrophic forgetting for the assessment of mammographic breast density. More specifically, we propose a simple yet effective approach, adapting Elastic weight consolidation (EWC) using the global batch normalization (BN) statistics of the original dataset. The results of this study provide guidance for the deployment of clinical deep learning models where continuous learning is needed for domain expansion.

Wed Mar 24 2021
Machine Learning
Addressing catastrophic forgetting for medical domain expansion
Model brittleness is a key concern when deploying deep learning models. A model that has high performance at one institution may suffer a significant decline in performance when tested at other institutions. An alternative approach is to fine-tune the model on subsequent institutions after training on the original institution.
1
11
42
Wed Nov 25 2020
Machine Learning
Advancing diagnostic performance and clinical usability of neural networks via adversarial training and dual batch normalization
The study shows that adversarially trained models can significantly enhance the ability of pathology detection. The accuracy of such models was equal to standard models when sufficiently large datasets and dual batch norm training were used.
0
0
0
Wed Jun 24 2020
Machine Learning
Bayesian Sampling Bias Correction: Training with the Right Loss Function
Sampling bias causes large discrepancies between model performance in the lab and in more realistic settings. It is often overlooked at training time or addressed on an ad-hoc basis. We derive a family of loss functions to train models in the presence of sampling bias.
0
0
0
Wed Feb 10 2021
Computer Vision
Partial transfusion: on the expressive influence of trainable batch norm parameters for transfer learning
0
0
0
Thu Jan 23 2020
Machine Learning
A Multi-site Study of a Breast Density Deep Learning Model for Full-field Digital Mammography Images and Synthetic Mammography Images
A DL model was trained to predict BI-RADS breast density using FFDM images acquired from 2008 to 2017. The model demonstrated substantial agreement with the original reporting radiologists for all three datasets. With adaptation, performance improved for Site 2 using only 500 SM images from that site.
0
0
0
Wed Mar 31 2021
Artificial Intelligence
Spectral decoupling allows training transferable neural networks in medical imaging
Deep neural networks show impressive performance in medical imaging tasks. Many current networks generalise poorly to data unseen during training. Spectral decoupling encourages the neural network to learn more features.
1
0
0