Search

Background: Recent advancements in large language models (LLMs) offer potential benefits in healthcare, particularly in processing extensive patient records. However, existing benchmarks do not fully assess LLMs' capability in handling real-world, lengthy clinical data. Methods: We present the LongHealth benchmark, comprising 20 detailed fictional patient cases across various diseases, with each case containing 5,090 to 6,754 words. The benchmark challenges LLMs with 400 multiple-choice questions in three categories: information extraction, negation, and sorting, challenging LLMs to extract and interpret information from large clinical documents. Results: We evaluated nine open-source LLMs with a minimum of 16,000 tokens and also included OpenAI's proprietary and cost-efficient GPT-3.5 Turbo for comparison. The highest accuracy was observed for Mixtral-8x7B-Instruct-v0.1, particularly in tasks focused on information retrieval from single and multiple patient documents. However, all models struggled significantly in tasks requiring the identification of missing information, highlighting a critical area for improvement in clinical data interpretation. Conclusion: While LLMs show considerable potential for processing long clinical documents, their current accuracy levels are insufficient for reliable clinical use, especially in scenarios requiring the identification of missing information. The LongHealth benchmark provides a more realistic assessment of LLMs in a healthcare setting and highlights the need for further model refinement for safe and effective clinical application. We make the benchmark and evaluation code publicly available., Comment: 11 pages, 3 figures, 5 tables

This paper presents medBERTde, a pre-trained German BERT model specifically designed for the German medical domain. The model has been trained on a large corpus of 4.7 Million German medical documents and has been shown to achieve new state-of-the-art performance on eight different medical benchmarks covering a wide range of disciplines and medical document types. In addition to evaluating the overall performance of the model, this paper also conducts a more in-depth analysis of its capabilities. We investigate the impact of data deduplication on the model's performance, as well as the potential benefits of using more efficient tokenization methods. Our results indicate that domain-specific models such as medBERTde are particularly useful for longer texts, and that deduplication of training data does not necessarily lead to improved performance. Furthermore, we found that efficient tokenization plays only a minor role in improving model performance, and attribute most of the improved performance to the large amount of training data. To encourage further research, the pre-trained model weights and new benchmarks based on radiological data are made publicly available for use by the scientific community., Comment: Keno K. Bressem and Jens-Michalis Papaioannou and Paul Grundmann contributed equally

OBJECTIVES: Distant metastases (DM) are the primary driver of mortality for patients with early stage non-small cell lung cancer (NSCLC) receiving stereotactic body radiation therapy (SBRT), yet patient-level risk is difficult to predict. We developed and validated a model to predict individualized risk of DM in this population.METHODS: We utilized a multi-institutional database of 1,280 patients with cT1-3N0M0 NSCLC treated with SBRT from 2006-2015 for model development and internal validation. A Fine and Gray (FG) regression model was built to predict 1-year DM risk and compared to a random survival forests (RSF) model. The higher performing model was evaluated on an external dataset of 130 patients from a separate institution. Discriminatory performance was evaluated using the time-dependent area under the curve (AUC). Calibration was assessed graphically and with Brier scores.RESULTS: The FG model yielded an AUC of 0.71 (95% CI: 0.57-0.86) compared to the RSF's AUC of 0.69 [95% CI: 0.63-0.85] in the internal test set and was selected for further testing. On external validation, the FG model yielded an AUC 0.70 (95% CI: 0.57-0.83) with good calibration (Brier score 0.08). The model identified a high-risk patient subgroup with greater 1-year DM rates in the internal test (20.0% (3/15) v 2.9% (7/241), P=0.001) and external validation 21.4% (3/15) v 7.8% (9/116), P= 0.095). A model nomogram and online application was made available.CONCLUSIONS: We developed and externally validated a practical model that predicts DM risk in NSCLC patients receiving SBRT that may help select patients for systemic therapy.

Introduction: Immunotherapy-induced pneumonitis (IIP) is a serious side-effect which requires accurate diagnosis and management with high-dose corticosteroids. The differential diagnosis between IIP and other types of pneumonitis (OTP) remains challenging due to similar radiological patterns. This study was aimed to develop a prediction model to differentiate IIP from OTP in patients with stage IV non-small cell lung cancer (NSCLC) who developed pneumonitis during immunotherapy. Methods: Consecutive patients with metastatic NSCLC treated with immunotherapy in six centres in the Netherlands and Belgium from 2017 to 2020 were reviewed and cause-specific pneumonitis events were identified. Seven regions of interest (segmented lungs and spheroidal/cubical regions surrounding the inflammation) were examined to extract the most predictive radiomic features from the chest computed tomography images obtained at pneumonitis manifestation. Models were internally tested regarding discrimination, calibration and decisional benefit. To evaluate the clinical application of the models, predicted labels were compared with the separate clinical and radiological judgements. Results: A total of 556 patients were reviewed; 31 patients (5.6%) developed IIP and 41 patients developed OTP (7.4%). The line of immunotherapy was the only predictive factor in the clinical model (2nd versus 1st odds ratio = 0.08, 95% confidence interval:0.01–0.77). The best radiomic model was achieved using a 75-mm spheroidal region of interest which showed an optimism-corrected area under the receiver operating characteristic curve of 0.83 (95% confidence interval:0.77–0.95) with negative and positive predictive values of 80% and 79%, respectively. Good calibration and net benefits were achieved for the radiomic model across the entire range of probabilities. A correct diagnosis was provided by the radiomic model in 10 out of 12 cases with non-conclusive radiological judgements. Conclusion: Radiomic biomark

Public imaging datasets are critical for the development and evaluation of automated tools in cancer imaging. Unfortunately, many do not include annotations or image-derived features, complicating their downstream analysis. Artificial intelligence-based annotation tools have been shown to achieve acceptable performance and thus can be used to automatically annotate large datasets. As part of the effort to enrich public data available within NCI Imaging Data Commons (IDC), here we introduce AI-generated annotations for two collections of computed tomography images of the chest, NSCLC-Radiomics, and the National Lung Screening Trial. Using publicly available AI algorithms we derived volumetric annotations of thoracic organs at risk, their corresponding radiomics features, and slice-level annotations of anatomical landmarks and regions. The resulting annotations are publicly available within IDC, where the DICOM format is used to harmonize the data and achieve FAIR principles. The annotations are accompanied by cloud-enabled notebooks demonstrating their use. This study reinforces the need for large, publicly accessible curated datasets and demonstrates how AI can be used to aid in cancer imaging.

Recent advances in large language models (LLMs) have shown impressive ability in biomedical question-answering, but have not been adequately investigated for more specific biomedical applications. This study investigates the performance of LLMs such as the ChatGPT family of models (GPT-3.5s, GPT-4) in biomedical tasks beyond question-answering. Because no patient data can be passed to the OpenAI API public interface, we evaluated model performance with over 10000 samples as proxies for two fundamental tasks in the clinical domain - classification and reasoning. The first task is classifying whether statements of clinical and policy recommendations in scientific literature constitute health advice. The second task is causal relation detection from the biomedical literature. We compared LLMs with simpler models, such as bag-of-words (BoW) with logistic regression, and fine-tuned BioBERT models. Despite the excitement around viral ChatGPT, we found that fine-tuning for two fundamental NLP tasks remained the best strategy. The simple BoW model performed on par with the most complex LLM prompting. Prompt engineering required significant investment., Comment: 28 pages, 2 tables and 4 figures. Submitting for review

Radiotherapy (RT) toxicities can impair survival and quality-of-life, yet remain under-studied. Real-world evidence holds potential to improve our understanding of toxicities, but toxicity information is often only in clinical notes. We developed natural language processing (NLP) models to identify the presence and severity of esophagitis from notes of patients treated with thoracic RT. We fine-tuned statistical and pre-trained BERT-based models for three esophagitis classification tasks: Task 1) presence of esophagitis, Task 2) severe esophagitis or not, and Task 3) no esophagitis vs. grade 1 vs. grade 2-3. Transferability was tested on 345 notes from patients with esophageal cancer undergoing RT. Fine-tuning PubmedBERT yielded the best performance. The best macro-F1 was 0.92, 0.82, and 0.74 for Task 1, 2, and 3, respectively. Selecting the most informative note sections during fine-tuning improved macro-F1 by over 2% for all tasks. Silver-labeled data improved the macro-F1 by over 3% across all tasks. For the esophageal cancer notes, the best macro-F1 was 0.73, 0.74, and 0.65 for Task 1, 2, and 3, respectively, without additional fine-tuning. To our knowledge, this is the first effort to automatically extract esophagitis toxicity severity according to CTCAE guidelines from clinic notes. The promising performance provides proof-of-concept for NLP-based automated detailed toxicity monitoring in expanded domains., Comment: 17 pages, 6 tables, 1figure, submiting to JCO-CCI for review

Despite major advances in artificial intelligence (AI) for medicine and healthcare, the deployment and adoption of AI technologies remain limited in real-world clinical practice. In recent years, concerns have been raised about the technical, clinical, ethical and legal risks associated with medical AI. To increase real world adoption, it is essential that medical AI tools are trusted and accepted by patients, clinicians, health organisations and authorities. This work describes the FUTURE-AI guideline as the first international consensus framework for guiding the development and deployment of trustworthy AI tools in healthcare. The FUTURE-AI consortium was founded in 2021 and currently comprises 118 inter-disciplinary experts from 51 countries representing all continents, including AI scientists, clinicians, ethicists, and social scientists. Over a two-year period, the consortium defined guiding principles and best practices for trustworthy AI through an iterative process comprising an in-depth literature review, a modified Delphi survey, and online consensus meetings. The FUTURE-AI framework was established based on 6 guiding principles for trustworthy AI in healthcare, i.e. Fairness, Universality, Traceability, Usability, Robustness and Explainability. Through consensus, a set of 28 best practices were defined, addressing technical, clinical, legal and socio-ethical dimensions. The recommendations cover the entire lifecycle of medical AI, from design, development and validation to regulation, deployment, and monitoring. FUTURE-AI is a risk-informed, assumption-free guideline which provides a structured approach for constructing medical AI tools that will be trusted, deployed and adopted in real-world practice. Researchers are encouraged to take the recommendations into account in proof-of-concept stages to facilitate future translation towards clinical practice of medical AI.

Despite major advances in artificial intelligence (AI) for medicine and healthcare, the deployment and adoption of AI technologies remain limited in real-world clinical practice. In recent years, concerns have been raised about the technical, clinical, ethical and legal risks associated with medical AI. To increase real world adoption, it is essential that medical AI tools are trusted and accepted by patients, clinicians, health organisations and authorities. This work describes the FUTURE-AI guideline as the first international consensus framework for guiding the development and deployment of trustworthy AI tools in healthcare. The FUTURE-AI consortium was founded in 2021 and currently comprises 118 inter-disciplinary experts from 51 countries representing all continents, including AI scientists, clinicians, ethicists, and social scientists. Over a two-year period, the consortium defined guiding principles and best practices for trustworthy AI through an iterative process comprising an in-depth literature review, a modified Delphi survey, and online consensus meetings. The FUTURE-AI framework was established based on 6 guiding principles for trustworthy AI in healthcare, i.e. Fairness, Universality, Traceability, Usability, Robustness and Explainability. Through consensus, a set of 28 best practices were defined, addressing technical, clinical, legal and socio-ethical dimensions. The recommendations cover the entire lifecycle of medical AI, from design, development and validation to regulation, deployment, and monitoring. FUTURE-AI is a risk-informed, assumption-free guideline which provides a structured approach for constructing medical AI tools that will be trusted, deployed and adopted in real-world practice. Researchers are encouraged to take the recommendations into account in proof-of-concept stages to facilitate future translation towards clinical practice of medical AI.

Social determinants of health (SDoH) have an important impact on patient outcomes but are incompletely collected from the electronic health records (EHR). This study researched the ability of large language models to extract SDoH from free text in EHRs, where they are most commonly documented, and explored the role of synthetic clinical text for improving the extraction of these scarcely documented, yet extremely valuable, clinical data. 800 patient notes were annotated for SDoH categories, and several transformer-based models were evaluated. The study also experimented with synthetic data generation and assessed for algorithmic bias. Our best-performing models were fine-tuned Flan-T5 XL (macro-F1 0.71) for any SDoH, and Flan-T5 XXL (macro-F1 0.70). The benefit of augmenting fine-tuning with synthetic data varied across model architecture and size, with smaller Flan-T5 models (base and large) showing the greatest improvements in performance (delta F1 +0.12 to +0.23). Model performance was similar on the in-hospital system dataset but worse on the MIMIC-III dataset. Our best-performing fine-tuned models outperformed zero- and few-shot performance of ChatGPT-family models for both tasks. These fine-tuned models were less likely than ChatGPT to change their prediction when race/ethnicity and gender descriptors were added to the text, suggesting less algorithmic bias (p<0.05). At the patient-level, our models identified 93.8% of patients with adverse SDoH, while ICD-10 codes captured 2.0%. Our method can effectively extracted SDoH information from clinic notes, performing better compare to GPT zero- and few-shot settings. These models could enhance real-world evidence on SDoH and aid in identifying patients needing social support., Comment: Peer-reviewed version published at NPJ Digital Medicine: https://www.nature.com/articles/s41746-023-00970-0

Generative models such as DALL-E 2 could represent a promising future tool for image generation, augmentation, and manipulation for artificial intelligence research in radiology provided that these models have sufficient medical domain knowledge. Here we show that DALL-E 2 has learned relevant representations of X-ray images with promising capabilities in terms of zero-shot text-to-image generation of new images, continuation of an image beyond its original boundaries, or removal of elements, while pathology generation or CT, MRI, and ultrasound images are still limited. The use of generative models for augmenting and generating radiological data thus seems feasible, even if further fine-tuning and adaptation of these models to the respective domain is required beforehand., Comment: 4 Figures

Background Radiomic features may quantify characteristics present in medical imaging. However, the lack of standardized definitions and validated reference values have hampered clinical use. Purpose To standardize a set of 174 radiomic features. Materials and Methods Radiomic features were assessed in three phases. In phase I, 487 features were derived from the basic set of 174 features. Twenty-five research teams with unique radiomics software implementations computed feature values directly from a digital phantom, without any additional image processing. In phase II, 15 teams computed values for 1347 derived features using a CT image of a patient with lung cancer and predefined image processing configurations. In both phases, consensus among the teams on the validity of tentative reference values was measured through the frequency of the modal value and classified as follows: less than three matches, weak; three to five matches, moderate; six to nine matches, strong; 10 or more matches, very strong. In the final phase (phase III), a public data set of multimodality images (CT, fluorine 18 fluorodeoxyglucose PET, and T1-weighted MRI) from 51 patients with soft-tissue sarcoma was used to prospectively assess reproducibility of standardized features. Results Consensus on reference values was initially weak for 232 of 302 features (76.8%) at phase I and 703 of 1075 features (65.4%) at phase II. At the final iteration, weak consensus remained for only two of 487 features (0.4%) at phase I and 19 of 1347 features (1.4%) at phase II. Strong or better consensus was achieved for 463 of 487 features (95.1%) at phase I and 1220 of 1347 features (90.6%) at phase II. Overall, 169 of 174 features were standardized in the first two phases. In the final validation phase (phase III), most of the 169 standardized features could be excellently reproduced (166 with CT; 164 with PET; and 164 with MRI). Conclusion A set of 169 radiomics features was standardized, which enabled verifica

Recent advances in artificial intelligence research have led to a profusion of studies that apply deep learning to problems in image analysis and natural language processing among others. Additionally, the availability of open-source computational frameworks has lowered the barriers to implementing state-of-the-art methods across multiple domains. Albeit leading to major performance breakthroughs in some tasks, effective dissemination of deep learning algorithms remains challenging, inhibiting reproducibility and benchmarking studies, impeding further validation, and ultimately hindering their effectiveness in the cumulative scientific progress. In developing a platform for sharing research outputs, we present ModelHub.AI (www.modelhub.ai), a community-driven container-based software engine and platform for the structured dissemination of deep learning models. For contributors, the engine controls data flow throughout the inference cycle, while the contributor-facing standard template exposes model-specific functions including inference, as well as pre- and post-processing. Python and RESTful Application programming interfaces (APIs) enable users to interact with models hosted on ModelHub.AI and allows both researchers and developers to utilize models out-of-the-box. ModelHub.AI is domain-, data-, and framework-agnostic, catering to different workflows and contributors' preferences.

PURPOSE:To study the variability in volume change estimates of pulmonary nodules due to segmentation approaches used across several algorithms and to evaluate these effects on the ability to predict nodule malignancy. METHODS:We obtained 100 patient image datasets from the National Lung Screening Trial (NLST) that had a nodule detected on each of two consecutive low dose computed tomography (LDCT) scans, with an equal proportion of malignant and benign cases (50 malignant, 50 benign). Information about the nodule location for the cases was provided by a screen capture with a bounding box and its axial location was indicated. Five participating quantitative imaging network (QIN) institutions performed nodule segmentation using their preferred semi-automated algorithms with no manual correction; teams were allowed to provide additional manually corrected segmentations (analyzed separately). The teams were asked to provide segmentation masks for each nodule at both time points. From these masks, the volume was estimated for the nodule at each time point; the change in volume (absolute and percent change) across time points was estimated as well. We used the concordance correlation coefficient (CCC) to compare the similarity of computed nodule volumes (absolute and percent change) across algorithms. We used Logistic regression model on the change in volume (absolute change and percent change) of the nodules to predict the malignancy status, the area under the receiver operating characteristic curve (AUROC) and confidence intervals were reported. Because the size of nodules was expected to have a substantial effect on segmentation variability, analysis of change in volumes was stratified by lesion size, where lesions were grouped into those with a longest diameter of <8 mm and those with longest diameter ≥ 8 mm. RESULTS:We find that segmentation of the nodules shows substantial variability across algorithms, with the CCC ranging from 0.56 to 0.95 for change in volume

PurposeTo study the variability in volume change estimates of pulmonary nodules due to segmentation approaches used across several algorithms and to evaluate these effects on the ability to predict nodule malignancy.MethodsWe obtained 100 patient image datasets from the National Lung Screening Trial (NLST) that had a nodule detected on each of two consecutive low dose computed tomography (LDCT) scans, with an equal proportion of malignant and benign cases (50 malignant, 50 benign). Information about the nodule location for the cases was provided by a screen capture with a bounding box and its axial location was indicated. Five participating quantitative imaging network (QIN) institutions performed nodule segmentation using their preferred semi-automated algorithms with no manual correction; teams were allowed to provide additional manually corrected segmentations (analyzed separately). The teams were asked to provide segmentation masks for each nodule at both time points. From these masks, the volume was estimated for the nodule at each time point; the change in volume (absolute and percent change) across time points was estimated as well. We used the concordance correlation coefficient (CCC) to compare the similarity of computed nodule volumes (absolute and percent change) across algorithms. We used Logistic regression model on the change in volume (absolute change and percent change) of the nodules to predict the malignancy status, the area under the receiver operating characteristic curve (AUROC) and confidence intervals were reported. Because the size of nodules was expected to have a substantial effect on segmentation variability, analysis of change in volumes was stratified by lesion size, where lesions were grouped into those with a longest diameter of <8 mm and those with longest diameter ≥ 8 mm.ResultsWe find that segmentation of the nodules shows substantial variability across algorithms, with the CCC ranging from 0.56 to 0.95 for change

Artificial intelligence (AI) is emerging as a technology with the power to transform established industries, and with applications from automated manufacturing to advertising and facial recognition to fully autonomous transportation. Advances in each of these domains have led some to call AI the "fourth" industrial revolution [1]. In healthcare, AI is emerging as both a productive and disruptive force across many disciplines. This is perhaps most evident in Diagnostic Radiology and Pathology, specialties largely built around the processing and complex interpretation of medical images, where the role of AI is increasingly seen as both a boon and a threat. In Radiation Oncology as well, AI seems poised to reshape the specialty in significant ways, though the impact of AI has been relatively limited at present, and may rightly seem more distant to many, given the predominantly interpersonal and complex interventional nature of the specialty. In this overview, we will explore the current state and anticipated future impact of AI on Radiation Oncology, in detail, focusing on key topics from multiple stakeholder perspectives, as well as the role our specialty may play in helping to shape the future of AI within the larger spectrum of medicine.

In this study we assessed the repeatability of the values of radiomics features for small prostate tumors using test-retest Multiparametric Magnetic Resonance Imaging (mpMRI) images. The premise of radiomics is that quantitative image features can serve as biomarkers characterizing disease. For such biomarkers to be useful, repeatability is a basic requirement, meaning its value must remain stable between two scans, if the conditions remain stable. We investigated repeatability of radiomics features under various preprocessing and extraction configurations including various image normalization schemes, different image pre-filtering, 2D vs 3D texture computation, and different bin widths for image discretization. Image registration as means to re-identify regions of interest across time points was evaluated against human-expert segmented regions in both time points. Even though we found many radiomics features and preprocessing combinations with a high repeatability (Intraclass Correlation Coefficient (ICC) > 0.85), our results indicate that overall the repeatability is highly sensitive to the processing parameters (under certain configurations, it can be below 0.0). Image normalization, using a variety of approaches considered, did not result in consistent improvements in repeatability. There was also no consistent improvement of repeatability through the use of pre-filtering options, or by using image registration between timepoints to improve consistency of the region of interest localization. Based on these results we urge caution when interpreting radiomics features and advise paying close attention to the processing configuration details of reported results. Furthermore, we advocate reporting all processing details in radiomics studies and strongly recommend making the implementation available.