Clinical, Cultural, Computational, and Regulatory Considerations to Deploy AI in Radiology: Perspectives of RSNA and MICCAI Experts.

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. The Radiological Society of North of America (RSNA) and the Medical Image Computing and Computer Assisted Intervention (MICCAI) Society have led a series of joint panels and seminars focused on the present impact and future directions of artificial intelligence (AI) in radiology. These conversations have collected viewpoints from multidisciplinary experts in radiology, medical imaging, and machine learning on the current clinical penetration of AI technology in radiology, and how it is impacted by trust, reproducibility, explainability, and accountability. The collective points-both practical and philosophical-define the cultural changes for radiologists and AI scientists working together and describe the challenges ahead for AI technologies to meet broad approval. This article presents the perspectives of experts from MICCAI and RSNA on the clinical, cultural, computational, and regulatory considerations-coupled with recommended reading materials-essential to adopt AI technology successfully in radiology and more generally in clinical practice. The report emphasizes the importance of collaboration to improve clinical deployment and highlights the need to integrate clinical and medical imaging data and introduces strategies to ensure smooth and incentivized integration. ©RSNA, 2024.

From Bench to Bedside With Large Language Models: AJR Expert Panel Narrative Review.

Large language models (LLMs) hold immense potential to revolutionize radiology. However, their integration into practice requires careful consideration. Artificial intelligence (AI) chatbots and general-purpose LLMs have potential pitfalls related to privacy, transparency, and accuracy, limiting their current clinical readiness. Thus, LLM-based tools must be optimized for radiology practice to overcome these limitations. While research and validation for radiology applications remain in their infancy, commercial products incorporating LLMs are becoming available alongside promises of transforming practice. To help radiologists navigate this landscape, this AJR Expert Panel Narrative Review provides a multidimensional perspective on LLMs, encompassing considerations from bench (development and optimization) to bedside (use in practice). At present, LLMs are not autonomous entities that can replace expert decision-making, and radiologists remain responsible for the content of their reports. Patient-facing tools, particularly medical AI chatbots, require additional guardrails to ensure safety and prevent misuse. Still, if responsibly implemented, LLMs are well-positioned to transform efficiency and quality in radiology. Radiologists must be well-informed and proactively involved in guiding the implementation of LLMs in practice to mitigate risks and maximize benefits to patient care.

The new era of artificial intelligence in neuroradiology: current research and promising tools.

Radiology has a number of characteristics that make it an especially suitable medical discipline for early artificial intelligence (AI) adoption. These include having a well-established digital workflow, standardized protocols for image storage, and numerous well-defined interpretive activities. The more than 200 commercial radiologic AI-based products recently approved by the Food and Drug Administration (FDA) to assist radiologists in a number of narrow image-analysis tasks such as image enhancement, workflow triage, and quantification, corroborate this observation. However, in order to leverage AI to boost efficacy and efficiency, and to overcome substantial obstacles to widespread successful clinical use of these products, radiologists should become familiarized with the emerging applications in their particular areas of expertise. In light of this, in this article we survey the existing literature on the application of AI-based techniques in neuroradiology, focusing on conditions such as vascular diseases, epilepsy, and demyelinating and neurodegenerative conditions. We also introduce some of the algorithms behind the applications, briefly discuss a few of the challenges of generalization in the use of AI models in neuroradiology, and skate over the most relevant commercially available solutions adopted in clinical practice. If well designed, AI algorithms have the potential to radically improve radiology, strengthening image analysis, enhancing the value of quantitative imaging techniques, and mitigating diagnostic errors.

A radiologia tem uma série de características que a torna uma disciplina médica especialmente adequada à adoção precoce da inteligência artificial (IA), incluindo um fluxo de trabalho digital bem estabelecido, protocolos padronizados para armazenamento de imagens e inúmeras atividades interpretativas bem definidas. Tal adequação é corroborada pelos mais de 200 produtos radiológicos comerciais baseados em IA recentemente aprovados pelo Food and Drug Administration (FDA) para auxiliar os radiologistas em uma série de tarefas restritas de análise de imagens, como quantificação, triagem de fluxo de trabalho e aprimoramento da qualidade das imagens. Entretanto, para o aumento da eficácia e eficiência da IA, além de uma utilização clínica bem-sucedida dos produtos que utilizam essa tecnologia, os radiologistas devem estar atualizados com as aplicações em suas áreas específicas de atuação. Assim, neste artigo, pesquisamos na literatura existente aplicações baseadas em IA em neurorradiologia, mais especificamente em condições como doenças vasculares, epilepsia, condições desmielinizantes e neurodegenerativas. Também abordamos os principais algoritmos por trás de tais aplicações, discutimos alguns dos desafios na generalização no uso desses modelos e introduzimos as soluções comercialmente disponíveis mais relevantes adotadas na prática clínica. Se cautelosamente desenvolvidos, os algoritmos de IA têm o potencial de melhorar radicalmente a radiologia, aperfeiçoando a análise de imagens, aumentando o valor das técnicas de imagem quantitativas e mitigando erros de diagnóstico.

Lessons Learned in Building Expertly Annotated Multi-Institution Datasets and Hosting the RSNA AI Challenges.

The Radiological Society of North America (RSNA) has held artificial intelligence competitions to tackle real-world medical imaging problems at least annually since 2017. This article examines the challenges and processes involved in organizing these competitions, with a specific emphasis on the creation and curation of high-quality datasets. The collection of diverse and representative medical imaging data involves dealing with issues of patient privacy and data security. Furthermore, ensuring quality and consistency in data, which includes expert labeling and accounting for various patient and imaging characteristics, necessitates substantial planning and resources. Overcoming these obstacles requires meticulous project management and adherence to strict timelines. The article also highlights the potential of crowdsourced annotation to progress medical imaging research. Through the RSNA competitions, an effective global engagement has been realized, resulting in innovative solutions to complex medical imaging problems, thus potentially transforming health care by enhancing diagnostic accuracy and patient outcomes. Keywords: Use of AI in Education, Artificial Intelligence © RSNA, 2024.

Checklist for Reproducibility of Deep Learning in Medical Imaging.

The application of deep learning (DL) in medicine introduces transformative tools with the potential to enhance prognosis, diagnosis, and treatment planning. However, ensuring transparent documentation is essential for researchers to enhance reproducibility and refine techniques. Our study addresses the unique challenges presented by DL in medical imaging by developing a comprehensive checklist using the Delphi method to enhance reproducibility and reliability in this dynamic field. We compiled a preliminary checklist based on a comprehensive review of existing checklists and relevant literature. A panel of 11 experts in medical imaging and DL assessed these items using Likert scales, with two survey rounds to refine responses and gauge consensus. We also employed the content validity ratio with a cutoff of 0.59 to determine item face and content validity. Round 1 included a 27-item questionnaire, with 12 items demonstrating high consensus for face and content validity that were then left out of round 2. Round 2 involved refining the checklist, resulting in an additional 17 items. In the last round, 3 items were deemed non-essential or infeasible, while 2 newly suggested items received unanimous agreement for inclusion, resulting in a final 26-item DL model reporting checklist derived from the Delphi process. The 26-item checklist facilitates the reproducible reporting of DL tools and enables scientists to replicate the study's results.

Performance of ChatGPT on the Brazilian Radiology and Diagnostic Imaging and Mammography Board Examinations.

This prospective exploratory study conducted from January 2023 through May 2023 evaluated the ability of ChatGPT to answer questions from Brazilian radiology board examinations, exploring how different prompt strategies can influence performance using GPT-3.5 and GPT-4. Three multiple-choice board examinations that did not include image-based questions were evaluated: (a) radiology and diagnostic imaging, (b) mammography, and (c) neuroradiology. Five different styles of zero-shot prompting were tested: (a) raw question, (b) brief instruction, (c) long instruction, (d) chain-of-thought, and (e) question-specific automatic prompt generation (QAPG). The QAPG and brief instruction prompt strategies performed best for all examinations (P < .05), obtaining passing scores (≥60%) on the radiology and diagnostic imaging examination when testing both versions of ChatGPT. The QAPG style achieved a score of 60% for the mammography examination using GPT-3.5 and 76% using GPT-4. GPT-4 achieved a score up to 65% in the neuroradiology examination. The long instruction style consistently underperformed, implying that excessive detail might harm performance. GPT-4's scores were less sensitive to prompt style changes. The QAPG prompt style showed a high volume of the "A" option but no statistical difference, suggesting bias was found. GPT-4 passed all three radiology board examinations, and GPT-3.5 passed two of three examinations when using an optimal prompt style. Keywords: ChatGPT, Artificial Intelligence, Board Examinations, Radiology and Diagnostic Imaging, Mammography, Neuroradiology © RSNA, 2023 See also the commentary by Trivedi and Gichoya in this issue.

Performance of the Winning Algorithms of the RSNA 2022 Cervical Spine Fracture Detection Challenge.

Purpose To evaluate and report the performance of the winning algorithms of the Radiological Society of North America Cervical Spine Fracture AI Challenge. Materials and Methods The competition was open to the public on Kaggle from July 28 to October 27, 2022. A sample of 3112 CT scans with and without cervical spine fractures (CSFx) were assembled from multiple sites (12 institutions across six continents) and prepared for the competition. The test set had 1093 scans (private test set: n = 789; mean age, 53.40 years ± 22.86 [SD]; 509 males; public test set: n = 304; mean age, 52.51 years ± 20.73; 189 males) and 847 fractures. The eight top-performing artificial intelligence (AI) algorithms were retrospectively evaluated, and the area under the receiver operating characteristic curve (AUC) value, F1 score, sensitivity, and specificity were calculated. Results A total of 1108 contestants composing 883 teams worldwide participated in the competition. The top eight AI models showed high performance, with a mean AUC value of 0.96 (95% CI: 0.95, 0.96), mean F1 score of 90% (95% CI: 90%, 91%), mean sensitivity of 88% (95% Cl: 86%, 90%), and mean specificity of 94% (95% CI: 93%, 96%). The highest values reported for previous models were an AUC of 0.85, F1 score of 81%, sensitivity of 76%, and specificity of 97%. Conclusion The competition successfully facilitated the development of AI models that could detect and localize CSFx on CT scans with high performance outcomes, which appear to exceed known values of previously reported models. Further study is needed to evaluate the generalizability of these models in a clinical environment. Keywords: Cervical Spine, Fracture Detection, Machine Learning, Artificial Intelligence Algorithms, CT, Head/Neck Supplemental material is available for this article. © RSNA, 2024.

Assessing robustness and generalization of a deep neural network for brain MS lesion segmentation on real-world data.

OBJECTIVES: Evaluate the performance of a deep learning (DL)-based model for multiple sclerosis (MS) lesion segmentation and compare it to other DL and non-DL algorithms. METHODS: This ambispective, multicenter study assessed the performance of a DL-based model for MS lesion segmentation and compared it to alternative DL- and non-DL-based methods. Models were tested on internal (n = 20) and external (n = 18) datasets from Latin America, and on an external dataset from Europe (n = 49). We also examined robustness by rescanning six patients (n = 6) from our MS clinical cohort. Moreover, we studied inter-human annotator agreement and discussed our findings in light of these results. Performance and robustness were assessed using intraclass correlation coefficient (ICC), Dice coefficient (DC), and coefficient of variation (CV). RESULTS: Inter-human ICC ranged from 0.89 to 0.95, while spatial agreement among annotators showed a median DC of 0.63. Using expert manual segmentations as ground truth, our DL model achieved a median DC of 0.73 on the internal, 0.66 on the external, and 0.70 on the challenge datasets. The performance of our DL model exceeded that of the alternative algorithms on all datasets. In the robustness experiment, our DL model also achieved higher DC (ranging from 0.82 to 0.90) and lower CV (ranging from 0.7 to 7.9%) when compared to the alternative methods. CONCLUSION: Our DL-based model outperformed alternative methods for brain MS lesion segmentation. The model also proved to generalize well on unseen data and has a robust performance and low processing times both on real-world and challenge-based data. CLINICAL RELEVANCE STATEMENT: Our DL-based model demonstrated superior performance in accurately segmenting brain MS lesions compared to alternative methods, indicating its potential for clinical application with improved accuracy, robustness, and efficiency. KEY POINTS: • Automated lesion load quantification in MS patients is valuable; however, more accurate methods are still necessary. • A novel deep learning model outperformed alternative MS lesion segmentation methods on multisite datasets. • Deep learning models are particularly suitable for MS lesion segmentation in clinical scenarios.

Performance of RAPID noncontrast CT stroke platform in large vessel occlusion and intracranial hemorrhage detection.

Background: Noncontrast CT (NCCT) is used to evaluate for intracerebral hemorrhage (ICH) and ischemia in acute ischemic stroke (AIS). Large vessel occlusions (LVOs) are a major cause of AIS, but challenging to detect on NCCT. Aims: The purpose of this study is to evaluate an AI software called RAPID NCCT Stroke (RAPID, iSchemaView, Menlo Park, CA) for ICH and LVO detection compared to expert readers. Methods: In this IRB approved retrospective, multicenter study, stand-alone performance of the software was assessed based on the consensus of 3 neuroradiologists and sensitivity and specificity were determined. The platform's performance was then compared to interpretation by readers comprised of eight general radiologists (GR) and three neuroradiologists (NR) in detecting ICH and hyperdense vessel sign (HVS) indicating LVO. Results: A total of 244 cases were included. Of the 244, 115 were LVOs and 26 were ICHs. One hundred three cases did not have LVO nor ICH. Stand-alone performance of the software demonstrated sensitivities and specificities of 96.2 and 99.5% for ICH and 63.5 and 95.1% for LVO detection. Compared to all 11 readers and eight GR readers only respectively, the software demonstrated superiority, achieving significantly higher sensitivities (63.5% versus 43.6%, p < 0.0001 and 63.5% versus 40.9%, p = 0.001). Conclusion: The RAPID NCCT Stroke platform demonstrates superior performance to radiologists for detecting LVO from a NCCT. Use of this software platform could lead to earlier LVO detection and expedited transfer of these patients to a thrombectomy capable center.

Decreasing false-positive detection of intracranial hemorrhage (ICH) using RAPID ICH 3.

INTRODUCTION: The prompt detection of intracranial hemorrhage (ICH) on a non-contrast head CT (NCCT) is critical for the appropriate triage of patients, particularly in high volume/high acuity settings. Several automated ICH detection tools have been introduced; however, at present, most suffer from suboptimal specificity leading to false-positive notifications. METHODS: NCCT scans from 4 large databases were evaluated for the presence of an ICH (IPH, IVH, SAH or SDH) of >0.4 ml using fully-automated RAPID ICH 3.0 as compared to consensus detection from at least two neuroradiology experts. Scans were excluded for (1) severe CT artifacts, (2) prior neurosurgical procedures, or (3) recent intravenous contrast. ICH detection accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and positive and negative likelihood ratios by were determined. RESULTS: A total of 881 studies were included. The automated software correctly identified 453/463 ICH-positive cases and 416/418 ICH-negative cases, resulting in a sensitivity of 97.84% and specificity 99.52%, positive predictive value 99.56%, and negative predictive value 97.65% for ICH detection. The positive and negative likelihood ratios for ICH detection were similarly favorable at 204.49 and 0.02 respectively. Mean processing time was <40 seconds. CONCLUSIONS: In this large data set of nearly 900 patients, the automated software demonstrated high sensitivity and specificity for ICH detection, with rare false-positives.

The RSNA Cervical Spine Fracture CT Dataset.

This dataset is composed of cervical spine CT images with annotations related to fractures; it is available at https://www.kaggle.com/competitions/rsna-2022-cervical-spine-fracture-detection/.

Validation of a deep learning algorithm for bone age estimation among patients in the city of São Paulo, Brazil / Validação de algoritmo de aprendizado profundo para detecção da idade óssea em pacientes de São Paulo, Brasil

Abstract Objective: To validate a deep learning (DL) model for bone age estimation in individuals in the city of São Paulo, comparing it with the Greulich and Pyle method. Materials and Methods: This was a cross-sectional study of hand and wrist radiographs obtained for the determination of bone age. The manual analysis was performed by an experienced radiologist. The model used was based on a convolutional neural network that placed third in the 2017 Radiological Society of North America challenge. The mean absolute error (MAE) and the root-mean-square error (RMSE) were calculated for the model versus the radiologist, with comparisons by sex, race, and age. Results: The sample comprised 714 examinations. There was a correlation between the two methods, with a coefficient of determination of 0.94. The MAE of the predictions was 7.68 months, and the RMSE was 10.27 months. There were no statistically significant differences between sexes or among races (p > 0.05). The algorithm overestimated bone age in younger individuals (p = 0.001). Conclusion: Our DL algorithm demonstrated potential for estimating bone age in individuals in the city of São Paulo, regardless of sex and race. However, improvements are needed, particularly in relation to its use in younger patients.

Resumo Objetivo: Validar em indivíduos paulistas um modelo de aprendizado profundo (deep learning - DL) para estimativa da idade óssea, comparando-o com o método de Greulich e Pyle. Materiais e Métodos: Estudo transversal com radiografias de mão e punho para idade óssea. A análise manual foi feita por um radiologista experiente. Foi usado um modelo baseado em uma rede neural convolucional que ficou em terceiro lugar no desafio de 2017 da Radiological Society of North America. Calcularam-se o erro médio absoluto (mean absolute error - MAE) e a raiz do erro médio quadrado (root mean-square error - RMSE) do modelo contra o radiologista, com comparações entre sexo, etnia e idade. Resultados: A amostra compreendia 714 exames. Houve correlação entre ambos os métodos com coeficiente de determinação de 0,94. O MAE das predições foi 7,68 meses e a RMSE foi 10,27 meses. Não houve diferenças estatisticamente significantes entre sexos ou raças (p > 0,05). O algoritmo superestimou a idade óssea nos mais jovens (p = 0,001). Conclusão: O nosso algoritmo de DL demonstrou potencial para estimar a idade óssea em indivíduos paulistas, independentemente do sexo e da raça. Entretanto, há necessidade de aprimoramentos, particularmente em pacientes mais jovens.

Reproducibility of Deep Learning Algorithms Developed for Medical Imaging Analysis: A Systematic Review.

Since 2000, there have been more than 8000 publications on radiology artificial intelligence (AI). AI breakthroughs allow complex tasks to be automated and even performed beyond human capabilities. However, the lack of details on the methods and algorithm code undercuts its scientific value. Many science subfields have recently faced a reproducibility crisis, eroding trust in processes and results, and influencing the rise in retractions of scientific papers. For the same reasons, conducting research in deep learning (DL) also requires reproducibility. Although several valuable manuscript checklists for AI in medical imaging exist, they are not focused specifically on reproducibility. In this study, we conducted a systematic review of recently published papers in the field of DL to evaluate if the description of their methodology could allow the reproducibility of their findings. We focused on the Journal of Digital Imaging (JDI), a specialized journal that publishes papers on AI and medical imaging. We used the keyword "Deep Learning" and collected the articles published between January 2020 and January 2022. We screened all the articles and included the ones which reported the development of a DL tool in medical imaging. We extracted the reported details about the dataset, data handling steps, data splitting, model details, and performance metrics of each included article. We found 148 articles. Eighty were included after screening for articles that reported developing a DL model for medical image analysis. Five studies have made their code publicly available, and 35 studies have utilized publicly available datasets. We provided figures to show the ratio and absolute count of reported items from included studies. According to our cross-sectional study, in JDI publications on DL in medical imaging, authors infrequently report the key elements of their study to make it reproducible.

Artificial Intelligence in Radiology: A Private Practice Perspective From a Large Health System in Latin America.

In the field of radiology, the use of artificial intelligence (AI) is increasing. Even though healthcare facilities are interested in using this technology, having success with an AI project can be challenging. There is a myriad of AI solutions today, and comparing them can be challenging. Moreover, the implementation process involves alignment with many different areas. In our institution, we have been testing, developing, deploying, and monitoring AI solutions for the last four years. This article intends to share our experience and highlight the most important points to ensure a successful project based on our experience in the setting of a large private practice in Latin America.

Non-task expert physicians benefit from correct explainable AI advice when reviewing X-rays.

Artificial intelligence (AI)-generated clinical advice is becoming more prevalent in healthcare. However, the impact of AI-generated advice on physicians' decision-making is underexplored. In this study, physicians received X-rays with correct diagnostic advice and were asked to make a diagnosis, rate the advice's quality, and judge their own confidence. We manipulated whether the advice came with or without a visual annotation on the X-rays, and whether it was labeled as coming from an AI or a human radiologist. Overall, receiving annotated advice from an AI resulted in the highest diagnostic accuracy. Physicians rated the quality of AI advice higher than human advice. We did not find a strong effect of either manipulation on participants' confidence. The magnitude of the effects varied between task experts and non-task experts, with the latter benefiting considerably from correct explainable AI advice. These findings raise important considerations for the deployment of diagnostic advice in healthcare.