Pub Date : 2024-08-07DOI: 10.1101/2024.08.06.24311530
Sabrina Sghirripa, Gaurav Bhalerao, Ludovica Griffanti, Grace Gillis, Clare E Mackay, Natalie Voets, Stephanie Wong, Mark Jenkinson
Given the relationship between hippocampal atrophy and cognitive impairment in various pathological conditions, hippocampus segmentation from MRI is an important task in neuroimaging. Manual segmentation, though considered the gold standard, is time-consuming and error-prone, leading to the development of numerous automatic segmentation methods. However, no study has yet independently compared the performance of traditional, deep learning-based, and hippocampal subfield segmentation methods within a single investigation. We evaluated nine automatic hippocampal segmentation methods (FreeSurfer, FastSurfer, FIRST, e2dhipseg, HippMapper, Hippodeep, FreeSurfer-Subfields, HippUnfold and HSF) across three datasets with manually segmented hippocampus labels. Performance metrics included overlap with manual labels, correlations between manual and automatic volumes, diagnostic group differentiation, and systematically located false positives and negatives. Most methods, especially deep learning-based ones, performed well on public datasets but showed more error and variability on unseen data. Many methods tended to over-segment, particularly at the anterior hippocampus border, but were able to distinguish between healthy controls, MCI, and dementia patients based on hippocampal volume. Our findings highlight the challenges in hippocampal segmentation from MRI and the need for more publicly accessible datasets with manual labels across diverse ages and pathological conditions.
{"title":"Evaluating Traditional, Deep Learning, and Subfield Methods for Automatically Segmenting the Hippocampus from MRI","authors":"Sabrina Sghirripa, Gaurav Bhalerao, Ludovica Griffanti, Grace Gillis, Clare E Mackay, Natalie Voets, Stephanie Wong, Mark Jenkinson","doi":"10.1101/2024.08.06.24311530","DOIUrl":"https://doi.org/10.1101/2024.08.06.24311530","url":null,"abstract":"Given the relationship between hippocampal atrophy and cognitive impairment in various pathological conditions, hippocampus segmentation from MRI is an important task in neuroimaging. Manual segmentation, though considered the gold standard, is time-consuming and error-prone, leading to the development of numerous automatic segmentation methods. However, no study has yet independently compared the performance of traditional, deep learning-based, and hippocampal subfield segmentation methods within a single investigation. We evaluated nine automatic hippocampal segmentation methods (FreeSurfer, FastSurfer, FIRST, e2dhipseg, HippMapper, Hippodeep, FreeSurfer-Subfields, HippUnfold and HSF) across three datasets with manually segmented hippocampus labels. Performance metrics included overlap with manual labels, correlations between manual and automatic volumes, diagnostic group differentiation, and systematically located false positives and negatives. Most methods, especially deep learning-based ones, performed well on public datasets but showed more error and variability on unseen data. Many methods tended to over-segment, particularly at the anterior hippocampus border, but were able to distinguish between healthy controls, MCI, and dementia patients based on hippocampal volume. Our findings highlight the challenges in hippocampal segmentation from MRI and the need for more publicly accessible datasets with manual labels across diverse ages and pathological conditions.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"39 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141941564","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-06DOI: 10.1101/2024.08.06.24311513
frederic De Beukelaer, Sophie De Beukelaer, Laura Wuyts, Mohammed El Halal, Martin Wiesmann, Hani Ridwan, Charlotte S. Weyland
BACKGROUND AND PURPOSE Neuroimaging of intracranial vessels with implanted stents (ICS) and flowdiverters (FD) is limited by artifacts. Photon-Counting-Detector-Computed Tomography (PCD-CT) is characterized by a higher resolution. The purpose of this study was to assess the image quality of ultra-high-resolution (UHR) PCD-CT-Angiography (PCD-CTA) and spectral reconstructions to define the best imaging parameters for the evaluation of vessel visibility in ICS and FD. MATERIALS AND METHODS Retrospective analysis of consecutive patients with implanted ICS or FD, who received a PCD-CTA between April 2023 and March 2024. Polyenergetic (PE), virtual monoenergetic imaging (VMI), pure lumen (PL) and iodine (I) reconstructions with different kiloelectron volt (keV) levels (keV 40, 60 and 80) and reconstruction kernels (Body vascular kernel (Bv) 48, Bv56, Bv64, Bv72, Bv76) were acquired to evaluate image quality and assessed by 2 independent radiologists using a 5-point Likert scale and regions of interest (ROI). The different kernels, keV and the optimized spectral reconstructions were compared in descriptive analysis. RESULTS In total, 12 patients with 9 FDs and 6 ICSs were analyzed. In terms of quantitative image quality, sharper kernels as Bv64 and Bv72 yielded increased image noise, and decreased signal to noise (SNR) and contrast to noise ratio (CNR) compared to the smoothest kernel Bv48, (p<0.01). Among the different keV levels and kernels, readers selected the 40 keV level (p<0.01) and sharper kernels (in the majority of cases Bv72) as the best to visualize the in-stent vessel lumen. Assessing the different spectral reconstructions virtual monoenergetic and iodine reconstructions proved to be best to evaluate in-stent vessel lumen (p<0.01). CONCLUSIONS Our preliminary study suggests that PCD-CTA and spectral reconstructions with sharper reconstruction kernels and a low keV level of 40 seem to be beneficial to achieve optimal image quality for the evaluation of ICS and FD. Iodine and virtual monoenergetic reconstructions were superior to pure lumen and polyenergetic reconstructions to evaluate in-stent vessel lumen
{"title":"Photon-Counting Detector Computed Tomography Angiography to assess intracranial stents and flow diverters: in vivo study comprising ultra-high resolution spectral reconstructions.","authors":"frederic De Beukelaer, Sophie De Beukelaer, Laura Wuyts, Mohammed El Halal, Martin Wiesmann, Hani Ridwan, Charlotte S. Weyland","doi":"10.1101/2024.08.06.24311513","DOIUrl":"https://doi.org/10.1101/2024.08.06.24311513","url":null,"abstract":"BACKGROUND AND PURPOSE Neuroimaging of intracranial vessels with implanted stents (ICS) and flowdiverters (FD) is limited by artifacts. Photon-Counting-Detector-Computed Tomography (PCD-CT) is characterized by a higher resolution. The purpose of this study was to assess the image quality of ultra-high-resolution (UHR) PCD-CT-Angiography (PCD-CTA) and spectral reconstructions to define the best imaging parameters for the evaluation of vessel visibility in ICS and FD.\u0000MATERIALS AND METHODS Retrospective analysis of consecutive patients with implanted ICS or FD, who received a PCD-CTA between April 2023 and March 2024. Polyenergetic (PE), virtual monoenergetic imaging (VMI), pure lumen (PL) and iodine (I) reconstructions with different kiloelectron volt (keV) levels (keV 40, 60 and 80) and reconstruction kernels (Body vascular kernel (Bv) 48, Bv56, Bv64, Bv72, Bv76) were acquired to evaluate image quality and assessed by 2 independent radiologists using a 5-point Likert scale and regions of interest (ROI). The different kernels, keV and the optimized spectral reconstructions were compared in descriptive analysis.\u0000RESULTS In total, 12 patients with 9 FDs and 6 ICSs were analyzed. In terms of quantitative image quality, sharper kernels as Bv64 and Bv72 yielded increased image noise, and decreased signal to noise (SNR) and contrast to noise ratio (CNR) compared to the smoothest kernel Bv48, (p<0.01). Among the different keV levels and kernels, readers selected the 40 keV level (p<0.01) and sharper kernels (in the majority of cases Bv72) as the best to visualize the in-stent vessel lumen. Assessing the different spectral reconstructions virtual monoenergetic and iodine reconstructions proved to be best to evaluate in-stent vessel lumen (p<0.01). CONCLUSIONS Our preliminary study suggests that PCD-CTA and spectral reconstructions with sharper reconstruction kernels and a low keV level of 40 seem to be beneficial to achieve optimal image quality for the evaluation of ICS and FD. Iodine and virtual monoenergetic reconstructions were superior to pure lumen and polyenergetic reconstructions to evaluate in-stent vessel lumen","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"5 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141941566","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-04DOI: 10.1101/2024.08.02.24311402
Grace Gillis, Gaurav Bhalerao, Jasmine Blane, Robert Mitchell, Pieter M Pretorius, Celeste McCracken, Thomas E Nichols, Stephen M Smith, Karla L Miller, Fidel Alfaro-Almagro, Vanessa Raymont, Lola Martos, Clare E Mackay, Ludovica Griffanti
Introduction The analysis tools and statistical methods used in large neuroimaging research studies differ from those applied in clinical contexts, making it unclear whether these techniques can be translated to a memory clinic setting. The Oxford Brain Health Clinic (OBHC) was established in 2020 to bridge this gap between research studies and memory clinics.
{"title":"From Big Data to the clinic: methodological and statistical enhancements to implement the UK Biobank imaging framework in a memory clinic","authors":"Grace Gillis, Gaurav Bhalerao, Jasmine Blane, Robert Mitchell, Pieter M Pretorius, Celeste McCracken, Thomas E Nichols, Stephen M Smith, Karla L Miller, Fidel Alfaro-Almagro, Vanessa Raymont, Lola Martos, Clare E Mackay, Ludovica Griffanti","doi":"10.1101/2024.08.02.24311402","DOIUrl":"https://doi.org/10.1101/2024.08.02.24311402","url":null,"abstract":"<strong>Introduction</strong> The analysis tools and statistical methods used in large neuroimaging research studies differ from those applied in clinical contexts, making it unclear whether these techniques can be translated to a memory clinic setting. The Oxford Brain Health Clinic (OBHC) was established in 2020 to bridge this gap between research studies and memory clinics.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"22 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141941569","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-03DOI: 10.1101/2024.08.02.24311385
Jessica Y Im, Neghemi Micah, Amy E Perkins, Kai Mei, Michael Geagan, Peter B Noël
All in-vivo medical imaging is impacted by patient motion, especially respiratory motion, which has a significant influence on clinical workflows in diagnostic imaging and radiation therapy. Many technologies such as motion artifact reduction and tumor tracking algorithms have been developed to compensate for respiratory motion during imaging. To assess these technologies, respiratory motion phantoms (RMPs) are required as preclinical testing environments, for instance, in computed tomography (CT). However, current RMPs are highly simplified and do not exhibit realistic tissue structures or deformation patterns. With the rise of more complex motion compensation technologies such as deep learning-based algorithms, there is a need for more realistic RMPs. This work introduces PixelPrint4D, a 3D printing method designed to fabricate lifelike, patient-specific deformable lung phantoms for CT imaging. The phantom demonstrated accurate replication of patient lung structures, textures, and attenuation profiles. Furthermore, it exhibited accurate nonrigid deformations, volume changes, and attenuation changes under compression. PixelPrint4D enables the production of highly realistic RMPs, surpassing existing models to offer more robust testing environments for a diverse array of novel CT technologies.
{"title":"PixelPrint4D: A 3D printing method of fabricating patient-specific deformable CT phantoms for respiratory motion applications.","authors":"Jessica Y Im, Neghemi Micah, Amy E Perkins, Kai Mei, Michael Geagan, Peter B Noël","doi":"10.1101/2024.08.02.24311385","DOIUrl":"https://doi.org/10.1101/2024.08.02.24311385","url":null,"abstract":"All in-vivo medical imaging is impacted by patient motion, especially respiratory motion, which has a significant influence on clinical workflows in diagnostic imaging and radiation therapy. Many technologies such as motion artifact reduction and tumor tracking algorithms have been developed to compensate for respiratory motion during imaging. To assess these technologies, respiratory motion phantoms (RMPs) are required as preclinical testing environments, for instance, in computed tomography (CT). However, current RMPs are highly simplified and do not exhibit realistic tissue structures or deformation patterns. With the rise of more complex motion compensation technologies such as deep learning-based algorithms, there is a need for more realistic RMPs. This work introduces PixelPrint4D, a 3D printing method designed to fabricate lifelike, patient-specific deformable lung phantoms for CT imaging. The phantom demonstrated accurate replication of patient lung structures, textures, and attenuation profiles. Furthermore, it exhibited accurate nonrigid deformations, volume changes, and attenuation changes under compression. PixelPrint4D enables the production of highly realistic RMPs, surpassing existing models to offer more robust testing environments for a diverse array of novel CT technologies.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"64 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141941568","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-03DOI: 10.1101/2024.08.02.24311396
Alex Mirugwe, Lillian Tamale, Juwa Nyirenda
Introduction: Tuberculosis remains a significant global health challenge, necessitating more efficient and accurate diagnostic methods. Methods: This study evaluates the performance of various convolutional neural network (CNN) architectures VGG16, VGG19, ResNet50, ResNet101, ResNet152, and Inception-ResNet-V2 in classifying chest X-ray (CXR) images as either normal or TB positive. The dataset comprised 4,200 CXR images, with 700 labeled as TB-positive and 3,500 as normal. We also examined the impact of data augmentation on model performance and analyzed the training times and the number of parameters for each architecture. Results: Our results showed that VGG16 outperformed the other models across all evaluation metrics, achieving an accuracy of 99.4%, precision of 97.9%, recall of 98.6%, F1-score of 98.3%, and AUC-ROC of 98.25%. Surprisingly, data augmentation did not improve performance, suggesting that the original dataset's diversity was sufficient. Furthermore, models with large numbers of parameters, such as ResNet152 and Inception-ResNet-V2, required longer training times without yielding proportionally better performance. Discussion: These findings highlight the importance of selecting the appropriate model architecture based on task-specific requirements. While more complex models with larger parameter counts may seem advantageous, they do not necessarily offer superior performance and often come with increased computational costs. Conclusion: The study demonstrates the potential of simpler models such as VGG16 to effectively diagnose TB from CXR images, providing a balance between performance and computational efficiency. This insight can guide future research and practical implementations in medical image classification.
{"title":"Improving Tuberculosis Detection in Chest X-ray Images through Transfer Learning and Deep Learning: A Comparative Study of CNN Architectures","authors":"Alex Mirugwe, Lillian Tamale, Juwa Nyirenda","doi":"10.1101/2024.08.02.24311396","DOIUrl":"https://doi.org/10.1101/2024.08.02.24311396","url":null,"abstract":"Introduction: Tuberculosis remains a significant global health challenge, necessitating more efficient and accurate diagnostic methods.\u0000Methods: This study evaluates the performance of various convolutional neural network (CNN) architectures VGG16, VGG19, ResNet50, ResNet101, ResNet152, and Inception-ResNet-V2 in classifying chest X-ray (CXR) images as either normal or TB positive. The dataset comprised 4,200 CXR images, with 700 labeled as TB-positive and 3,500 as normal. We also examined the impact of data augmentation on model performance and analyzed the training times and the number of parameters for each architecture.\u0000Results: Our results showed that VGG16 outperformed the other models across all evaluation metrics, achieving an accuracy of 99.4%, precision of 97.9%, recall of 98.6%, F1-score of 98.3%, and AUC-ROC of 98.25%. Surprisingly, data augmentation did not improve performance, suggesting that the original dataset's diversity was sufficient. Furthermore, models with large numbers of parameters, such as ResNet152 and Inception-ResNet-V2, required longer training times without yielding proportionally better performance.\u0000Discussion: These findings highlight the importance of selecting the appropriate model architecture based on task-specific requirements. While more complex models with larger parameter counts may seem advantageous, they do not necessarily offer superior performance and often come with increased computational costs.\u0000Conclusion: The study demonstrates the potential of simpler models such as VGG16 to effectively diagnose TB from CXR images, providing a balance between performance and computational efficiency. This insight can guide future research and practical implementations in medical image classification.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"10 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141941571","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-03DOI: 10.1101/2024.08.02.24311394
Diana L. Giraldo, Hamza Khan, Gustavo Pineda, Zhihua Liang, Alfonso Lozano Castillo, Bart Van Wijmeersch, Henry Woodruff, Philippe Lambin, Eduardo Romero, Liesbet M. Peeters, Jan Sijbers
Magnetic resonance imaging (MRI) is crucial for diagnosing and monitoring of multiple sclerosis (MS) as it is used to assess lesions in the brain and spinal cord. However, in real-world clinical settings, MRI scans are often acquired with thick slices, limiting their utility for automated quantitative analyses. This work presents a single-image super-resolution (SR) reconstruction framework that leverages SR convolutional neural networks (CNN) to enhance the through-plane resolution of structural MRI in people with MS (PwMS). Our strategy involves the supervised fine-tuning of CNN architectures, guided by a content loss function that promotes perceptual quality, as well as reconstruction accuracy, to recover high-level image features. Extensive evaluation with MRI data of PwMS shows that our SR strategy leads to more accurate MRI reconstructions than competing methods. Furthermore, it improves lesion segmentation on low-resolution MRI, approaching the performance achievable with high-resolution images. Results demonstrate the potential of our SR framework to facilitate the use of low-resolution retrospective MRI from real-world clinical settings to investigate quantitative image-based biomarkers of MS.
磁共振成像（MRI）是诊断和监测多发性硬化症（MS）的关键，因为它可用于评估大脑和脊髓的病变。然而，在现实世界的临床环境中，磁共振成像扫描通常采用厚切片采集，限制了其在自动定量分析中的应用。这项研究提出了一种单图像超分辨率（SR）重建框架，利用 SR 卷积神经网络（CNN）来提高多发性硬化症（PwMS）患者结构性 MRI 的通面分辨率。我们的策略包括在内容损失函数的指导下对 CNN 架构进行有监督的微调，以提高感知质量和重建准确性，从而恢复高级图像特征。使用 PwMS 核磁共振数据进行的广泛评估表明，与其他竞争方法相比，我们的 SR 策略能带来更准确的核磁共振重建。此外，它还改善了低分辨率核磁共振成像的病灶分割，接近高分辨率图像所能达到的性能。研究结果表明，我们的 SR 框架有潜力促进低分辨率回顾性 MRI 在实际临床环境中的应用，从而研究基于图像的 MS 定量生物标记物。
{"title":"Perceptual super-resolution in multiple sclerosis MRI","authors":"Diana L. Giraldo, Hamza Khan, Gustavo Pineda, Zhihua Liang, Alfonso Lozano Castillo, Bart Van Wijmeersch, Henry Woodruff, Philippe Lambin, Eduardo Romero, Liesbet M. Peeters, Jan Sijbers","doi":"10.1101/2024.08.02.24311394","DOIUrl":"https://doi.org/10.1101/2024.08.02.24311394","url":null,"abstract":"Magnetic resonance imaging (MRI) is crucial for diagnosing and monitoring of multiple sclerosis (MS) as it is used to assess lesions in the brain and spinal cord. However, in real-world clinical settings, MRI scans are often acquired with thick slices, limiting their utility for automated quantitative analyses. This work presents a single-image super-resolution (SR) reconstruction framework that leverages SR convolutional neural networks (CNN) to enhance the through-plane resolution of structural MRI in people with MS (PwMS). Our strategy involves the supervised fine-tuning of CNN architectures, guided by a content loss function that promotes perceptual quality, as well as reconstruction accuracy, to recover high-level image features. Extensive evaluation with MRI data of PwMS shows that our SR strategy leads to more accurate MRI reconstructions than competing methods. Furthermore, it improves lesion segmentation on low-resolution MRI, approaching the performance achievable with high-resolution images. Results demonstrate the potential of our SR framework to facilitate the use of low-resolution retrospective MRI from real-world clinical settings to investigate quantitative image-based biomarkers of MS.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"166 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141941637","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-03DOI: 10.1101/2024.08.02.24311223
Ludovica Griffanti, Florence Serres, Laura Cini, Jessica Walsh, Taylor Hanayik, Usama Pervaiz, Stephen Smith, Heidi Johansen-Berg, James Rose, Mamta Bajre
With the rise in numbers of people living with dementia and new disease modifying therapies entering the market, there is increasing need for brain magnetic resonance imaging (MRI) for diagnosis and safety monitoring. The number of scans that need reporting is expected to rapidly grow. Clinical radiology reports are currently largely qualitative and variable in structure and content. By contrast, research software typically uses automated methods to extract quantitative metrics from brain scans. To better understand the unmet clinical need for brain reporting software for dementia we conducted a barrier to adoption study using the Lean Assessment Process (LAP)methodology. We first assessed the role of brain imaging in the diagnostic pathway for people with suspected dementia in the NHS in England. We then explored the views of (neuro)radiologists, neurologists and psychiatrists on the potential benefits and level of acceptance of software to support brain MRI analysis, using the FMRIB software library (FSL) as a technology exemplar. The main perceived utilities of the proposed software were: increased diagnostic confidence; support for delivery of disease modifying therapies; and the possibility to compare individual results with population norms. In addition to assessment of global atrophy, hippocampal atrophy and white matter hyperintensities, additional user requirements included assessment of microbleeds, segmentation of multiple brain structures, clear information about the control population used for reference, and possibility to compare multiple scans. The main barriers to adoption related to the limited availability of 3T MRI scanners in the UK, integration into the clinical workflow, and the need to demonstrate cost-effectiveness. These findings will guide future technical development, clinical validation, and health economic evaluation.
{"title":"Brain magnetic resonance imaging software to support dementia diagnosis in routine clinical practice: a barrier to adoption study in the National Health Service (NHS) England","authors":"Ludovica Griffanti, Florence Serres, Laura Cini, Jessica Walsh, Taylor Hanayik, Usama Pervaiz, Stephen Smith, Heidi Johansen-Berg, James Rose, Mamta Bajre","doi":"10.1101/2024.08.02.24311223","DOIUrl":"https://doi.org/10.1101/2024.08.02.24311223","url":null,"abstract":"With the rise in numbers of people living with dementia and new disease modifying therapies entering the market, there is increasing need for brain magnetic resonance imaging (MRI) for diagnosis and safety monitoring. The number of scans that need reporting is expected to rapidly grow. Clinical radiology reports are currently largely qualitative and variable in structure and content. By contrast, research software typically uses automated methods to extract quantitative metrics from brain scans.\u0000To better understand the unmet clinical need for brain reporting software for dementia we conducted a barrier to adoption study using the Lean Assessment Process (LAP)methodology. We first assessed the role of brain imaging in the diagnostic pathway for people with suspected dementia in the NHS in England. We then explored the views of (neuro)radiologists, neurologists and psychiatrists on the potential benefits and level of acceptance of software to support brain MRI analysis, using the FMRIB software library (FSL) as a technology exemplar.\u0000The main perceived utilities of the proposed software were: increased diagnostic confidence; support for delivery of disease modifying therapies; and the possibility to compare individual results with population norms. In addition to assessment of global atrophy, hippocampal atrophy and white matter hyperintensities, additional user requirements included assessment of microbleeds, segmentation of multiple brain structures, clear information about the control population used for reference, and possibility to compare multiple scans. The main barriers to adoption related to the limited availability of 3T MRI scanners in the UK, integration into the clinical workflow, and the need to demonstrate cost-effectiveness. These findings will guide future technical development, clinical validation, and health economic evaluation.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"10 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141941610","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-02DOI: 10.1101/2024.07.31.24311123
Kazufumi Suzuki, Kayoko Abe, Shuji Sakai
Purpose: This study aimed to evaluate the potential of GPT-4, a large language model, in assisting radiologists to determine brain magnetic resonance imaging (MRI) protocols. Materials and methods: We used brain MRI protocols from a specific hospital, covering 20 diseases or examination purposes, excluding brain tumor protocols. GPT-4 was given system prompts to add one MRI sequence for the basic brain MRI protocol and disease names were input as user prompts. The model's suggestions were evaluated by two radiologists with over 20 years of relevant experience. Suggestions were scored based on their alignment with the hospital's protocol as follows: 0 for inappropriate, 1 for acceptable but nonmatching, and 2 for matching the protocol. The experiment was conducted in both Japanese and English to compare GPT-4's performance in different languages. Results: GPT-4 scored 27/40 points in English and 28/40 points in Japanese. GPT-4 gave inappropriate suggestions for Moyamoya disease and neuromyelitis optica in both languages and cerebral infarction in Japanese. For the other protocols, the suggested sequences were either appropriate or better. The suggestions in English differed from those in Japanese for seven protocols. Conclusion: GPT-4 generally suggested suitable MRI sequences. The study demonstrates that GPT-4 can help radiologists to determine protocols; however, further research is required for the radiological application of LLMs.
{"title":"Can GPT-4 suggest the optimal sequence for brain magnetic resonance imaging?","authors":"Kazufumi Suzuki, Kayoko Abe, Shuji Sakai","doi":"10.1101/2024.07.31.24311123","DOIUrl":"https://doi.org/10.1101/2024.07.31.24311123","url":null,"abstract":"Purpose: This study aimed to evaluate the potential of GPT-4, a large language model, in assisting radiologists to determine brain magnetic resonance imaging (MRI) protocols.\u0000Materials and methods: We used brain MRI protocols from a specific hospital, covering 20 diseases or examination purposes, excluding brain tumor protocols. GPT-4 was given system prompts to add one MRI sequence for the basic brain MRI protocol and disease names were input as user prompts. The model's suggestions were evaluated by two radiologists with over 20 years of relevant experience. Suggestions were scored based on their alignment with the hospital's protocol as follows: 0 for inappropriate, 1 for acceptable but nonmatching, and 2 for matching the protocol. The experiment was conducted in both Japanese and English to compare GPT-4's performance in different languages.\u0000Results: GPT-4 scored 27/40 points in English and 28/40 points in Japanese. GPT-4 gave inappropriate suggestions for Moyamoya disease and neuromyelitis optica in both languages and cerebral infarction in Japanese. For the other protocols, the suggested sequences were either appropriate or better. The suggestions in English differed from those in Japanese for seven protocols. Conclusion: GPT-4 generally suggested suitable MRI sequences. The study demonstrates that GPT-4 can help radiologists to determine protocols; however, further research is required for the radiological application of LLMs.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"366 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141887004","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-02DOI: 10.1101/2024.08.01.24311342
Eloise S Ockenden, Simon Mpooya, J. Alison Noble, Goylette F Chami
Liver diseases are a leading cause of death worldwide, with an estimated 2 million deaths each year. Causes of liver disease are diffi- cult to ascertain, especially in sub-Saharan Africa where there is a high prevalence of infectious diseases such as hepatitis B and schistosomi- asis, along with alcohol use. Point-of-care ultrasound often is used in low-resource settings for diagnosis of liver disease due to its portabil- ity and low cost. For classification models that can automatically stage liver disease from ultrasound video, the region of interest is liver tissue. A fully-automated pipeline for liver tissue identification in ultrasound video is presented. Ultrasound video data was collected using a low-cost, portable ultrasound machine in rural areas of Uganda. The pipeline first detects the diaphragm in each ultrasound video frame, then segments the diaphragm to ultimately use this segmentation to infer the position of liver tissue in each frame. This pipeline outperforms directly segmenting liver tissue with an intersection over union of 0.83 compared to 0.62. This pipeline also shows improved results with respect to the ease of clinical interpretation and anticipated clinical utility.
{"title":"An Automated Pipeline for the Identification of Liver Tissue in Ultrasound Video","authors":"Eloise S Ockenden, Simon Mpooya, J. Alison Noble, Goylette F Chami","doi":"10.1101/2024.08.01.24311342","DOIUrl":"https://doi.org/10.1101/2024.08.01.24311342","url":null,"abstract":"Liver diseases are a leading cause of death worldwide, with an estimated 2 million deaths each year. Causes of liver disease are diffi- cult to ascertain, especially in sub-Saharan Africa where there is a high prevalence of infectious diseases such as hepatitis B and schistosomi- asis, along with alcohol use. Point-of-care ultrasound often is used in low-resource settings for diagnosis of liver disease due to its portabil- ity and low cost. For classification models that can automatically stage liver disease from ultrasound video, the region of interest is liver tissue. A fully-automated pipeline for liver tissue identification in ultrasound video is presented. Ultrasound video data was collected using a low-cost, portable ultrasound machine in rural areas of Uganda. The pipeline first detects the diaphragm in each ultrasound video frame, then segments the diaphragm to ultimately use this segmentation to infer the position of liver tissue in each frame. This pipeline outperforms directly segmenting liver tissue with an intersection over union of 0.83 compared to 0.62. This pipeline also shows improved results with respect to the ease of clinical interpretation and anticipated clinical utility.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"44 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141881633","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-01DOI: 10.1101/2024.07.31.24311257
Rocío García-Mojón, Fernando Martín-Rodríguez, Mónica Fernández-Barciela
In this paper a study about breast cancer detection is presented. Mammography images in DICOM format are processed using Convolutional Neural Networks (CNNs) to get a pre-diagnosis. Of course, this preliminary result needs to be checked by a trained radiologist. CNNs are trained and checked using a big database that is publicly available. Standard measurements for success are computed (accuracy, precision, recall) obtaining outstanding results better than other examples from the literature.
本文介绍了一项关于乳腺癌检测的研究。使用卷积神经网络 (CNN) 处理 DICOM 格式的乳腺 X 射线图像，以获得预诊断结果。当然，这一初步结果需要由训练有素的放射科医生进行检查。CNN 通过一个公开的大型数据库进行训练和检查。通过计算成功的标准衡量标准（准确度、精确度、召回率），获得了优于其他文献实例的出色结果。
{"title":"Helping Breast Cancer Diagnosis on Mammographies using Convolutional Neural Networks","authors":"Rocío García-Mojón, Fernando Martín-Rodríguez, Mónica Fernández-Barciela","doi":"10.1101/2024.07.31.24311257","DOIUrl":"https://doi.org/10.1101/2024.07.31.24311257","url":null,"abstract":"In this paper a study about breast cancer detection is presented. Mammography images in DICOM format are processed using Convolutional Neural Networks (CNNs) to get a pre-diagnosis. Of course, this preliminary result needs to be checked by a trained radiologist. CNNs are trained and checked using a big database that is publicly available. Standard measurements for success are computed (accuracy, precision, recall) obtaining outstanding results better than other examples from the literature.","PeriodicalId":501358,"journal":{"name":"medRxiv - Radiology and Imaging","volume":"76 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141868494","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：