What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to improve the model performance and interpretability in the medical imaging protocol assignment task. Specifically, the research aims to evaluate the performance of different pre - trained BERT models (such as BERT, BioBERT, ClinicalBERT and RoBERTa) in the neuroradiology protocol assignment task and gain an in - depth understanding of the decision - making processes of these models. ### Decomposition of the Main Problem 1. **Improvement of Model Performance**: - Researchers hope to improve the accuracy of the medical imaging protocol classification task by fine - tuning the pre - trained BERT models. - They selected four pre - trained models: BERT, BioBERT, ClinicalBERT and RoBERTa, and fine - tuned them to adapt to the specific medical text classification task. 2. **Model Interpretability**: - In high - risk fields such as the medical environment, understanding the decision - making process of the model is crucial. - Researchers used the Integrated Gradients method to quantify the contribution of each word in the input text to the model's decision, and verified it by deleting important and unimportant words. - An experienced radiologist reviewed the word importance scores generated by the model to assess whether the model's decision was in line with human reasoning. 3. **Systematic Error Identification**: - Researchers analyzed the misclassification cases of the model and discovered potential systematic errors. - These errors may include multiple - choice questions, age - related results, ambiguous entries and obvious errors. ### Formula Representation To ensure that the formulas are correct and readable, the following are the Markdown - format representations of some key concepts involved in the paper: - **F1 Score**: An indicator used to measure the accuracy of the model, which combines precision and recall. The calculation formula is: \[ F1 = 2\times\frac{\text{Precision}\times\text{Recall}}{\text{Precision}+\text{Recall}} \] - **Integrated Gradients**: Used to calculate the importance of each word. The formula is as follows: \[ IG_i(x)=(x_i - x'_i)\cdot\int_{\alpha = 0}^{1}\frac{\partial F(x'+\alpha\cdot(x - x'))}{\partial x_i}d\alpha \] where \(x\) is the input text, \(x'\) is the baseline input (usually a zero - vector), and \(F\) is the model output. ### Conclusion The research results show that the fine - tuned BERT model exhibits performance close to the human level in the medical imaging protocol assignment task and can effectively identify key words. By detecting systematic errors, the research provides directions for improving the safety and practicality of the model. In addition, the interpretability of the model has also been enhanced, making its application in the clinical environment more reliable.

Exploring the performance and explainability of fine-tuned BERT models for neuroradiology protocol assignment

Evaluation of a BERT Natural Language Processing Model for Automating CT and MRI Triage and Protocol Selection

Automatic Assignment of Radiology Examination Protocols Using Pre-trained Language Models with Knowledge Distillation

Information extraction from weakly structured radiological reports with natural language queries

Clinical-BERT: Vision-Language Pre-training for Radiograph Diagnosis and Reports Generation.

BERT in Radiology: A Systematic Review of Natural Language Processing Applications

Why does my medical AI look at pictures of birds? Exploring the efficacy of transfer learning across domain boundaries

Highly accurate classification of chest radiographic reports using a deep learning natural language model pre-trained on 3.8 million text reports

Oversampling effect in pretraining for bidirectional encoder representations from transformers (BERT) to localize medical BERT and enhance biomedical BERT

Does the magic of BERT apply to medical code assignment? A quantitative study

Deep Learning-Based Natural Language Processing in Radiology: The Impact of Report Complexity, Disease Prevalence, Dataset Size, and Algorithm Type on Model Performance

Classification of Radiological Text in Small and Imbalanced Datasets in a Non-English Language

The Utility of General Domain Transfer Learning for Medical Language Tasks

Pre-training technique to localize medical BERT and enhance biomedical BERT

Disease Classification and Impact of Pretrained Deep Convolution Neural Networks on Diverse Medical Imaging Datasets across Imaging Modalities

Domain-Specific Language Model Pretraining for Biomedical Natural Language Processing

BioBERT: a pre-trained biomedical language representation model for biomedical text mining

Does Biomedical Training Lead to Better Medical Performance?

Fine-tuning language model embeddings to reveal domain knowledge: An explainable artificial intelligence perspective on medical decision making

Improved Fine-Tuning of In-Domain Transformer Model for Inferring COVID-19 Presence in Multi-Institutional Radiology Reports