What question did this study set out to answer?

To compare the classification performance and computational efficiency of various deep learning architectures on unstructured medical notes across varying levels of disease class imbalance.

July 2, 2022BMC Medical Research MethodologyOpen Access

A comparative study on deep learning models for text classification of unstructured medical notes with various levels of class imbalance

View Full Paper

Ask AI

Bookmark

Authors

HLHongxia LuUniversity of North Carolina at Chapel Hill LELouis EhwerhemuephaChildren's Hospital of Orange County CRCyril RakovskiChapman University

Discussion

Member takes

Implication

Comparative study demonstrates Transformer encoders excel in medical note classification across class imbalances, suggesting trade-offs between accuracy and computational efficiency.

Key Points

To compare the classification performance and computational efficiency of various deep learning architectures on unstructured medical notes across varying levels of disease class imbalance.
Evaluated seven deep learning architectures (CNN, Transformer encoder, pre-trained BERT-Base, RNN, GRU, LSTM, and Bi-LSTM) across 16 binary disease classification tasks using patient discharge summary notes.
Assessed model performance under different class imbalance ratios using AUC-ROC, AUC-PR, F1 score, balanced accuracy, and training runtimes.
Compared performance variations using three word embedding strategies: GloVe, BioWordVec, and no pre-trained embeddings.
The Transformer encoder achieved the highest classification performance across nearly all evaluated class imbalance scenarios.
For conditions with prevalence near or exceeding 50%, CNN achieved competitive accuracy while reducing training time by 91.3% compared to the Transformer encoder and 94.7% compared to BERT-Base.
BioWordVec embeddings marginally enhanced Bi-LSTM performance, CNN performed best without pre-trained embeddings, and GloVe embeddings reduced training times across all models.

Cite This Study

Lu et al. (2022) studied this question.

synapsesocial.com/papers/6a12e8e083732aa7db9e89e7 https://doi.org/10.1186/s12874-022-01665-y

View Full Paper

Ask AI

Bookmark

Implication

Comparative study demonstrates Transformer encoders excel in medical note classification across class imbalances, suggesting trade-offs between accuracy and computational efficiency.

A comparative study on deep learning models for text classification of unstructured medical notes with various levels of class imbalance

Discussion

Member takes

Implication

Key Points

Cite This Study

A comparative study on deep learning models for text classification of unstructured medical notes with various levels of class imbalance

Discussion

Member takes

Implication

Key Points

Cite This Study

Also Consider

Also Consider