Improved Fine-Tuning of In-Domain Transformer Model for Inferring COVID-19 Presence in Multi-Institutional Radiology Reports.
Ontology highlight
ABSTRACT:
SUBMITTER: Chambon P
PROVIDER: S-EPMC9629758 | biostudies-literature | 2022 Nov
REPOSITORIES: biostudies-literature
Publications
Improved Fine-Tuning of In-Domain Transformer Model for Inferring COVID-19 Presence in Multi-Institutional Radiology Reports.
Chambon Pierre P Cook Tessa S TS Langlotz Curtis P CP
Journal of digital imaging 20221102 1
Building a document-level classifier for COVID-19 on radiology reports could help assist providers in their daily clinical routine, as well as create large numbers of labels for computer vision models. We have developed such a classifier by fine-tuning a BERT-like model initialized from RadBERT, its continuous pre-training on radiology reports that can be used on all radiology-related tasks. RadBERT outperforms all biomedical pre-trainings on this COVID-19 task (P<0.01) and helps our fine-tuned ...[more]
Similar Datasets
Project description:BackgroundGiven the increasing number of people suffering from tinnitus, the accurate categorization of patients with actionable reports is attractive in assisting clinical decision making. However, this process requires experienced physicians and significant human labor. Natural language processing (NLP) has shown great potential in big data analytics of medical texts; yet, its application to domain-specific analysis of radiology reports is limited.ObjectiveThe aim of this study is to propose a novel approach in classifying actionable radiology reports of tinnitus patients using bidirectional encoder representations from transformer BERT-based models and evaluate the benefits of in domain pre-training (IDPT) along with a sequence adaptation strategy.MethodsA total of 5864 temporal bone computed tomography(CT) reports are labeled by two experienced radiologists as follows: (1) normal findings without notable lesions; (2) notable lesions but uncorrelated to tinnitus; and (3) at least one lesion considered as potential cause of tinnitus. We then constructed a framework consisting of deep learning (DL) neural networks and self-supervised BERT models. A tinnitus domain-specific corpus is used to pre-train the BERT model to further improve its embedding weights. In addition, we conducted an experiment to evaluate multiple groups of max sequence length settings in BERT to reduce the excessive quantity of calculations. After a comprehensive comparison of all metrics, we determined the most promising approach through the performance comparison of F1-scores and AUC values.ResultsIn the first experiment, the BERT finetune model achieved a more promising result (AUC-0.868, F1-0.760) compared with that of the Word2Vec-based models(AUC-0.767, F1-0.733) on validation data. In the second experiment, the BERT in-domain pre-training model (AUC-0.948, F1-0.841) performed significantly better than the BERT based model(AUC-0.868, F1-0.760). Additionally, in the variants of BERT fine-tuning models, Mengzi achieved the highest AUC of 0.878 (F1-0.764). Finally, we found that the BERT max-sequence-length of 128 tokens achieved an AUC of 0.866 (F1-0.736), which is almost equal to the BERT max-sequence-length of 512 tokens (AUC-0.868,F1-0.760).ConclusionIn conclusion, we developed a reliable BERT-based framework for tinnitus diagnosis from Chinese radiology reports, along with a sequence adaptation strategy to reduce computational resources while maintaining accuracy. The findings could provide a reference for NLP development in Chinese radiology reports.
| S-EPMC9338483 | biostudies-literature
Project description: Pulmonary nodules and nodule characteristics are important indicators of lung nodule malignancy. However, nodule information is often documented as free text in clinical narratives such as radiology reports in electronic health record systems. Natural language processing (NLP) is the key technology to extract and standardize patient information from radiology reports into structured data elements. This study aimed to develop an NLP system using state-of-the-art transformer models to extract pulmonary nodules and associated nodule characteristics from radiology reports. We identified a cohort of 3080 patients who underwent LDCT at the University of Florida health system and collected their radiology reports. We manually annotated 394 reports as the gold standard. We explored eight pretrained transformer models from three transformer architectures including bidirectional encoder representations from transformers (BERT), robustly optimized BERT approach (RoBERTa), and A Lite BERT (ALBERT), for clinical concept extraction, relation identification, and negation detection. We examined general transformer models pretrained using general English corpora, transformer models fine-tuned using a clinical corpus, and a large clinical transformer model, GatorTron, which was trained from scratch using 90 billion words of clinical text. We compared transformer models with two baseline models including a recurrent neural network implemented using bidirectional long short-term memory with a conditional random fields layer and support vector machines. RoBERTa-mimic achieved the best F1-score of 0.9279 for nodule concept and nodule characteristics extraction. ALBERT-base and GatorTron achieved the best F1-score of 0.9737 in linking nodule characteristics to pulmonary nodules. Seven out of eight transformers achieved the best F1-score of 1.0000 for negation detection. Our end-to-end system achieved an overall F1-score of 0.8869. This study demonstrated the advantage of state-of-the-art transformer models for pulmonary nodule information extraction from radiology reports.Supplementary informationThe online version contains supplementary material available at 10.1007/s41666-024-00166-5.
| S-EPMC11310180 | biostudies-literature
Project description:ObjectiveTo assess large language models on their ability to accurately infer cancer disease response from free-text radiology reports.Materials and methodsWe assembled 10 602 computed tomography reports from cancer patients seen at a single institution. All reports were classified into: no evidence of disease, partial response, stable disease, or progressive disease. We applied transformer models, a bidirectional long short-term memory model, a convolutional neural network model, and conventional machine learning methods to this task. Data augmentation using sentence permutation with consistency loss as well as prompt-based fine-tuning were used on the best-performing models. Models were validated on a hold-out test set and an external validation set based on Response Evaluation Criteria in Solid Tumors (RECIST) classifications.ResultsThe best-performing model was the GatorTron transformer which achieved an accuracy of 0.8916 on the test set and 0.8919 on the RECIST validation set. Data augmentation further improved the accuracy to 0.8976. Prompt-based fine-tuning did not further improve accuracy but was able to reduce the number of training reports to 500 while still achieving good performance.DiscussionThese models could be used by researchers to derive progression-free survival in large datasets. It may also serve as a decision support tool by providing clinicians an automated second opinion of disease response.ConclusionsLarge clinical language models demonstrate potential to infer cancer disease response from radiology reports at scale. Data augmentation techniques are useful to further improve performance. Prompt-based fine-tuning can significantly reduce the size of the training dataset.
| S-EPMC10531105 | biostudies-literature
Project description:Multivalent presentation of ligands often enhances receptor activation and downstream signaling. DNA origami offers precise nanoscale spacing of ligands, a potentially useful feature for therapeutic nanoparticles. Here, we use a square block DNA origami platform to explore the importance of spacing of CpG oligonucleotides. CpG engages Toll-like receptors and thereby acts to activate dendritic cells. Through in vitro cell-culture studies and in vivo tumor-treatment models, we demonstrate that square blocks induce Th1 immune polarization when CpG is spaced at 3.5 nm. We observe that this DNA origami vaccine enhances DC activation, antigen cross-presentation, CD8+ T cell activation, Th1-polarized CD4+ activation and natural killer cell activation. The vaccine also synergizes effectively with anti-PD-L1 for improved cancer immunotherapy in melanoma and lymphoma models and induces long-term T cell memory. Our results suggest that DNA origami may serve as a platform for controlling adjuvant spacing and co-delivering antigens in vaccines.
2024-02-01 | GSE251850 | GEO
Project description:Fine-tuning is an important technique in transfer learning that has achieved significant success in tasks that lack training data. However, as it is difficult to extract effective features for single-source domain fine-tuning when the data distribution difference between the source and the target domain is large, we propose a transfer learning framework based on multi-source domain called adaptive multi-source domain collaborative fine-tuning (AMCF) to address this issue. AMCF utilizes multiple source domain models for collaborative fine-tuning, thereby improving the feature extraction capability of model in the target task. Specifically, AMCF employs an adaptive multi-source domain layer selection strategy to customize appropriate layer fine-tuning schemes for the target task among multiple source domain models, aiming to extract more efficient features. Furthermore, a novel multi-source domain collaborative loss function is designed to facilitate the precise extraction of target data features by each source domain model. Simultaneously, it works towards minimizing the output difference among various source domain models, thereby enhancing the adaptability of the source domain model to the target data. In order to validate the effectiveness of AMCF, it is applied to seven public visual classification datasets commonly used in transfer learning, and compared with the most widely used single-source domain fine-tuning methods. Experimental results demonstrate that, in comparison with the existing fine-tuning methods, our method not only enhances the accuracy of feature extraction in the model but also provides precise layer fine-tuning schemes for the target task, thereby significantly improving the fine-tuning performance.
| S-EPMC11232598 | biostudies-literature
Project description:ObjectivesTo provide insights for on-site development of transformer-based structuring of free-text report databases by investigating different labeling and pre-training strategies.MethodsA total of 93,368 German chest X-ray reports from 20,912 intensive care unit (ICU) patients were included. Two labeling strategies were investigated to tag six findings of the attending radiologist. First, a system based on human-defined rules was applied for annotation of all reports (termed "silver labels"). Second, 18,000 reports were manually annotated in 197 h (termed "gold labels") of which 10% were used for testing. An on-site pre-trained model (Tmlm) using masked-language modeling (MLM) was compared to a public, medically pre-trained model (Tmed). Both models were fine-tuned on silver labels only, gold labels only, and first with silver and then gold labels (hybrid training) for text classification, using varying numbers (N: 500, 1000, 2000, 3500, 7000, 14,580) of gold labels. Macro-averaged F1-scores (MAF1) in percent were calculated with 95% confidence intervals (CI).ResultsTmlm,gold (95.5 [94.5-96.3]) showed significantly higher MAF1 than Tmed,silver (75.0 [73.4-76.5]) and Tmlm,silver (75.2 [73.6-76.7]), but not significantly higher MAF1 than Tmed,gold (94.7 [93.6-95.6]), Tmed,hybrid (94.9 [93.9-95.8]), and Tmlm,hybrid (95.2 [94.3-96.0]). When using 7000 or less gold-labeled reports, Tmlm,gold (N: 7000, 94.7 [93.5-95.7]) showed significantly higher MAF1 than Tmed,gold (N: 7000, 91.5 [90.0-92.8]). With at least 2000 gold-labeled reports, utilizing silver labels did not lead to significant improvement of Tmlm,hybrid (N: 2000, 91.8 [90.4-93.2]) over Tmlm,gold (N: 2000, 91.4 [89.9-92.8]).ConclusionsCustom pre-training of transformers and fine-tuning on manual annotations promises to be an efficient strategy to unlock report databases for data-driven medicine.Key points• On-site development of natural language processing methods that retrospectively unlock free-text databases of radiology clinics for data-driven medicine is of great interest. • For clinics seeking to develop methods on-site for retrospective structuring of a report database of a certain department, it remains unclear which of previously proposed strategies for labeling reports and pre-training models is the most appropriate in context of, e.g., available annotator time. • Using a custom pre-trained transformer model, along with a little annotation effort, promises to be an efficient way to retrospectively structure radiological databases, even if not millions of reports are available for pre-training.
| S-EPMC10181962 | biostudies-literature
Project description:BackgroundCommunity-engaged research (CEnR) involves institutions of higher education collaborating with organizations in their communities to exchange resources and knowledge to benefit a community's well-being. While community engagement is a critical aspect of a university's mission, tracking and reporting CEnR metrics can be challenging, particularly in terms of external community relations and federally funded research programs. In this study, we aimed to develop a method for classifying CEnR studies that have been submitted to our university's institutional review board (IRB) to capture the level of community involvement in research studies. Tracking studies in which communities are "highly engaged" enables institutions to obtain a more comprehensive understanding of the prevalence of CEnR.ObjectiveWe aimed to develop an updated experiment to classify CEnR and capture the distinct levels of involvement that a community partner has in the direction of a research study. To achieve this goal, we used a deep learning-based approach and evaluated the effectiveness of fine-tuning strategies on transformer-based models.MethodsIn this study, we used fine-tuning techniques such as discriminative learning rates and freezing layers to train and test 135 slightly modified classification models based on 3 transformer-based architectures: BERT (Bidirectional Encoder Representations from Transformers), Bio+ClinicalBERT, and XLM-RoBERTa. For the discriminative learning rate technique, we applied different learning rates to different layers of the model, with the aim of providing higher learning rates to layers that are more specialized to the task at hand. For the freezing layers technique, we compared models with different levels of layer freezing, starting with all layers frozen and gradually unfreezing different layer groups. We evaluated the performance of the trained models using a holdout data set to assess their generalizability.ResultsOf the models evaluated, Bio+ClinicalBERT performed particularly well, achieving an accuracy of 73.08% and an F1-score of 62.94% on the holdout data set. All the models trained in this study outperformed our previous models by 10%-23% in terms of both F1-score and accuracy.ConclusionsOur findings suggest that transfer learning is a viable method for tracking CEnR studies and provide evidence that the use of fine-tuning strategies significantly improves transformer-based models. Our study also presents a tool for categorizing the type and volume of community engagement in research, which may be useful in addressing the challenges associated with reporting CEnR metrics.
| S-EPMC9944122 | biostudies-literature
Project description:Traditional radiology reports are narrative texts that include a description of imaging findings. Recent implementation of advanced reporting software allows for incorporation of annotated key images and hyperlinks directly into text reports, but these tools usually do not substitute in-person consultations with radiologists, especially in challenging cases. Use of on-demand audio/visual reports with screen capture software is an emerging technology, providing a more engaged imaging service. Our study evaluates a video reporting tool that utilizes PACS integrated screen capture software for musculoskeletal imaging studies in the emergency department. Our hypothesis is that referring orthopedic surgeons would find that recorded audio/video reports add value to conventional reports, may increase engagement with radiology staff, and also facilitate understanding of imaging findings from urgent musculoskeletal cases. Seven radiologists prepared a total of 47 audiovisual reports for 9 attending orthopedic surgeons from the emergency department. We applied two surveys to evaluate the experience of the referring physicians using audio/visual reports as a complementary material from the conventional text report. Positive responses were statistically significant in most questions including: if the clinical suspicion was answered in the video; willingness to use such technology in other cases; if the audiovisual report made the imaging findings more understandable than the traditional report; and if the audiovisual report is faster to understand than the traditional text report. Use of audiovisual reports in emergency musculoskeletal cases is a new approach to evaluate potentially challenging cases. These results support the potential of this technology to re-establish the radiologist's role as an essential member of patient care and also provide more engaging, precise, and personalized reports. Further studies could streamline these methods in order to minimize work redundancy with traditional text reporting or even evaluate acceptance of using only audiovisual radiology reports. Additionally, widespread adoption would require integration with the entire radiology workflow including non-urgent cases and other medical specialties.
| S-EPMC6841833 | biostudies-literature
Project description:BackgroundThe purpose of this study was to assess the effectiveness of an Artificial Intelligence-Large Language Model (AI-LLM) at improving the readability of hand and wrist radiology reports.MethodsThe radiology reports of 100 hand and/or wrist radiographs, 100 hand and/or wrist computed tomography (CT) scans, and 100 hand and/or wrist magnetic resonance imaging (MRI) scans were extracted. The following prompt command was inserted into the AI-LLM: "Explain this radiology report to a patient in layman's terms in the second person: [Report Text]." The report length, Flesch reading ease score (FRES), and Flesch-Kincaid reading level (FKRL) were calculated for the original radiology report and the AI-LLM-generated report. The accuracy of the AI-LLM report was assessed via a 5-point Likert scale. Any "hallucination" produced by the AI-LLM-generated report was recorded.ResultsThere was a statistically significant improvement in mean FRES scores and FKRL scores in the AI-LLM-generated radiograph report, CT report, and MRI report. For all AI-LLM-generated reports, the mean reading level improved to below an eighth-grade reading level. The mean Likert score for the AI-LLM-generated radiograph report, CT report, and MRI report was 4.1 ± 0.6, 3.9 ± 0.6, and 3.9 ± 0.7, respectively. The hallucination rate in the AI-LLM-generated radiograph report, CT report, and MRI report was 3%, 6%, and 6%, respectively.ConclusionsThis study demonstrates that AI-LLM effectively improves the readability of hand and wrist radiology reports, underscoring the potential application of AI-LLM as a promising and innovative patient-centric strategy to improve patient comprehension of their imaging reports.Level of Evidence: IV.
| S-EPMC11574816 | biostudies-literature
Project description:Despite the reduction in turn-around times in radiology reporting with the use of speech recognition software, persistent communication errors can significantly impact the interpretation of radiology reports. Pre-filling a radiology report holds promise in mitigating reporting errors, and despite multiple efforts in literature to generate comprehensive medical reports, there lacks approaches that exploit the longitudinal nature of patient visit records in the MIMIC-CXR dataset. To address this gap, we propose to use longitudinal multi-modal data, i.e., previous patient visit CXR, current visit CXR, and the previous visit report, to pre-fill the "findings" section of the patient's current visit. We first gathered the longitudinal visit information for 26,625 patients from the MIMIC-CXR dataset, and created a new dataset called Longitudinal-MIMIC. With this new dataset, a transformer-based model was trained to capture the multi-modal longitudinal information from patient visit records (CXR images + reports) via a cross-attention-based multi-modal fusion module and a hierarchical memory-driven decoder. In contrast to previous works that only uses current visit data as input to train a model, our work exploits the longitudinal information available to pre-fill the "findings" section of radiology reports. Experiments show that our approach outperforms several recent approaches by ≥3% on F1 score, and ≥2% for BLEU-4, METEOR and ROUGE-L respectively. Code will be published at https://github.com/CelestialShine/Longitudinal-Chest-X-Ray.
| S-EPMC10947431 | biostudies-literature