Radiomics evaluation for the early detection of Alzheimer's dementia using T1-weighted MRI.

INTRODUCTION: Alzheimer's disease (AD), the most common cause of dementia, presents a global health crisis with its prevalence expected to triple worldwide by 2050, emphasizing the urgent need for early diagnosis to delay progression and improve patient quality of life. Our project aims to detect AD in its early phase by identifying subtle neuroanatomical changes with Radiomics features, offering a more accurate diagnosis. METHODS: The AssemblyNet segmentation model was used to analyze brain changes by employing anonymized T1 MRI scans from 416 patients. For each segmented label we extracted Radiomic features. After preprocessing of Radiomic features we trained four models, Gradient Booster, Random Forest, Support Vector Classifier, and XGBoost, in a 70%/20%/10% train, validation and test split. All models were hyperparameter tuned with GridSearch, Cross validation and evaluated with accuracy on the test data. RESULTS: 208 T1-weighted MRI scans were segmented, with 132 segmentation labels per patient, 1130 Radiomic features per segmentation, totalling in over 31 million features. For all four models we achieved accuracies between 0.71 and 0.86, and the machine learning model with highest accuracy were XGBoost, achieving an accuracy at 0.86 on the segmentation of the left inferior lateral ventricle. CONCLUSION: Our study's use of segmentation on T1-weighted MRI scans resulted promising accuracies for early AD diagnosis with the machine learning model XGBoost, peaking at 0.86 accuracy. Future research should aim to expand datasets and refine methodologies for broader applicability. IMPLICATION FOR PRACTICE: Implementing Radiomics for early AD detection using T1-weighted MRI scans could substantially improve diagnostic accuracy, enabling earlier interventions that may delay disease progression and improve outcomes, thereby requiring radiographers to adopt more advanced imaging techniques and analysis tools, as well as additional training to effectively interpret complex Radiomic data.

A Nordic survey on artificial intelligence in the radiography profession - Is the profession ready for a culture change?

INTRODUCTION: The impact of artificial intelligence (AI) on the radiography profession remains uncertain. Although AI has been increasingly used in clinical radiography, the perspectives of the radiography professionals in Nordic countries have yet to be examined. The primary aim was to examine views of Nordic radiographers 'on AI, with focus on perspectives, engagement, and knowledge of AI. METHODS: Radiographers from Denmark, Norway, Sweden, Iceland, Greenland, and the Faroe Island were invited through social media platforms to participate in an online survey from March to June 2023. The survey encompassed 29-items and included 4 sections a) demographics, b) barriers and enablers on AI, c) perspectives and experiences of AI and d) knowledge of AI in radiography. Edgars Schein's model of organizational culture was employed to analyse Nordic radiographers' perspectives on AI. RESULTS: Overall, a total of 421 respondents participated in the survey. A majority were positive/somewhat positive towards AI in radiography e.g., 77.9 % (n = 342) thought that AI would have a positive effect on the profession, and 26% thought that AI would reduce the administrative workload. Most radiographers agreed or strongly agreed that clinicians may have access to AI generated reports (76.8 %, n = 297). Nevertheless, a total of 86 (20.1%) agree or somewhat agreed that AI a potential risk for radiography. CONCLUSION: Nordic radiographers are generally positive towards AI, yet uncertainties regarding its implementation persist. The findings underscore the importance of understanding these challenges for the responsible integration of AI systems. Carefully weighing the expected influence of AI against key incentives will support a seamless integration of AI for the benefit not just of the patients, but also of the radiography profession. IMPLICATIONS FOR PRACTICE: Understanding incentives factors and barriers can help address uncertainties during implementation of AI in clinical practice.

Deep learning performance compared to healthcare experts in detecting wrist fractures from radiographs: A systematic review and meta-analysis.

OBJECTIVE: To perform a systematic review and meta-analysis of the diagnostic accuracy of deep learning (DL) algorithms in the diagnosis of wrist fractures (WF) on plain wrist radiographs, taking healthcare experts consensus as reference standard. METHODS: Embase, Medline, PubMed, Scopus and Web of Science were searched in the period from 1 Jan 2012 to 9 March 2023. Eligible studies were patients with wrist radiographs for radial and ulnar fractures as the target condition, studies using DL algorithms based on convolutional neural networks (CNN), and healthcare experts consensus as the minimum reference standard. Studies were assessed with a modified QUADAS-2 tool, and we applied a bivariate random-effects model for meta-analysis of diagnostic test accuracy data. RESULTS: Our study was registered at PROSPERO with ID: CRD42023431398. We included 6 unique studies for meta-analysis, with a total of 33,026 radiographs. CNN performance compared to reference standards for the included articles found a summary sensitivity of 92% (95% CI: 80%-97%) and a summary specificity of 93% (95% CI: 76%-98%). The generalized bivariate I-squared statistic indicated considerable heterogeneity between the studies (81.90%). Four studies had one or more domains at high risk of bias and two studies had concerns regarding applicability. CONCLUSION: The diagnostic accuracy of CNNs was comparable to that of healthcare experts in wrist radiographs for investigation of WF. There is a need for studies with a robust reference standard, external data-set validation and investigation of diagnostic performance of healthcare experts aided with CNNs. CLINICAL RELEVANCE STATEMENT: DL matches healthcare experts in diagnosing WFs, which potentially benefits patient diagnosis.

Nordic radiographers' and students' perspectives on artificial intelligence - A cross-sectional online survey.

INTRODUCTION: The integration of artificial intelligence (AI) into the domain of radiography holds substantial potential in various aspects including workflow efficiency, image processing, patient positioning, and quality assurance. The successful implementation of AI within a Radiology department necessitates the participation of key stakeholders, particularly radiographers. The study aimed to provide a comprehensive investigation about Nordic radiographers' perspectives and attitudes towards AI in radiography. METHODS: An online 29-item survey was distributed via social media platforms to Nordic students and radiographers working in Denmark, Norway, Sweden, Iceland, Greenland, and the Faroe Islands including items on demographics, specialization, educational background, place of work and perspectives and knowledge on AI. The items were a mix of closed-type and scaled questions, with the option for free-text responses when relevant. RESULTS: The survey received responses from all Nordic countries with 586 respondents, 26.8% males, 72.1% females, and 1.1% non-binary/self-defined or preferred not to say. The mean age was 37.2 with a standard deviation (SD) of ±12.1 years, and the mean number of years since qualification was 14.2 SD ± 10.3 years. A total of 43% (n = 254) of the respondents had not received any AI training in clinical practice. Whereas 13% (n = 76) had received AI during radiography undergrad training. A total of 77.9% (n = 412) expressed interest in pursuing AI education. The majority of respondents were aware of the potential use of AI (n = 485, 82.8%) and 39.1% (n = 204) had no reservations about AI. CONCLUSION: Overall, this study found that Nordic radiographers have a positive attitude toward AI. Very limited training or education has been provided to the radiographers. Especially since 82.8% reports on plans to implement AI in clinical practice. In general, awareness of AI applications is high, but the educational level is low for Nordic radiographers. IMPLICATION FOR PRACTICE: This study emphasises the favourable view of AI held by students and Nordic radiographers. However, there is a need for continuous professional development to facilitate the implementation and effective utilization of AI tools within the field of radiography.

Utilizing heat maps as explainable artificial intelligence for detecting abnormalities on wrist and elbow radiographs.

INTRODUCTION: Wrist and elbow radiographs, which plays a key role in diagnosing both fractures and degenerative conditions, present a diagnostic challenge due to intricate structures and subtle pathological signs. Artificial intelligence (AI) through deep learning models, has transformed diagnostic imaging, achieving accuracy rates, with explainable AI (XAI) and Gradient-weighted Class Activation Mapping (Grad-CAM) enhancing transparency of AI-driven diagnosis. METHOD: The MURA-dataset, a comprehensive collection of musculoskeletal radiographs, specifically focuses on wrist and elbow images, ensuring a spectrum of normal and abnormal conditions. An ensemble of transfer-learning models, including VGG16, VGG19, ResNet, DenseNet, InceptionV3 and Xception, was applied, with implemented Grad-CAM techniques, providing interpretable heat maps. The Dice Similarity Coefficient (DSC) evaluated the algorithm's efficiency in recognizing regions of interest. RESULTS: The average test accuracy of the 20 models were 0.81 (0.72-0.84), and 0.60 (0.49-0.73) for the wrist and elbow radiographs, respectively. The highest performing models were VGG16 with a test accuracy of 0.84, and DenseNet169 with a test accuracy of 0.73. The DSC were calculated for the six highest performing models, and agreements between algorithms were found on radiographs with metal, and only minimal agreement for radiographs with fractures. CONCLUSION: The study employed twenty transfer-learning models on wrist and elbow radiographs presenting accuracy and partial agreement with Grad-CAM technique evaluation. This study enables comprehension of model performance and avenues for potential enhancement. IMPLICATION FOR PRACTICE: The utilization of artificial intelligence, specifically transfer-learning models, could greatly enhance the accuracy and efficiency of diagnosing conditions from wrist and elbow radiographs. Additionally, the application of explainable AI techniques such as Grad-CAM can provide visual validation and transparency, thereby strengthening trust and adoption in clinical settings.

Collimation border with U-Net segmentation on chest radiographs compared to radiologists.

INTRODUCTION: Chest Radiography (CXR) is a common radiographic procedure. Radiation exposure to patients should be kept as low as reasonably achievable (ALARA), and monitored continuously as part of quality assurance (QA) programs. One of the most effective dose reduction tools is proper collimation practice. The purpose of this study is to determine whether a U-Net convolutional neural networks (U-CNN) can be trained to automatically segment the lungs and calculate an optimized collimation border on a limited CXR dataset. METHODS: 662 CXRs with manual lung segmentations were obtained from an open-source dataset. These were used to train and validate three different U-CNNs for automatic lung segmentation and optimal collimation. The U-CNN dimensions were 128 × 128, 256 × 256, and 512 × 512 pixels and validated with five-fold cross validation. The U-CNN with the highest area under the curve (AUC) was tested externally, using a dataset of 50 CXRs. Dice scores (DS) were used to compare U-CNN segmentations with manual segmentations by three radiographers and two junior radiologists. RESULTS: DS for the three U-CNN dimensions with segmentation of the lungs ranged from 0.93 to 0.96, respectively. DS of the collimation border for each U-CNN was 0.95 compared to the ground truth labels. DS for lung segmentation and collimation border between the junior radiologists was 0.97 and 0.97. One radiographer differed significantly from the U-CNN (p = 0.016). CONCLUSION: We demonstrated that a U-CNN could reliably segment the lungs and suggest a collimation border with great accuracy compared to junior radiologists. This algorithm has the potential to automate collimation auditing of CXRs. IMPLICATIONS FOR PRACTICE: Creating an automatic segmentation model of the lungs can produce a collimation border, which can be used in CXR QA programs.

The effect of Gaussian noise on pneumonia detection on chest radiographs, using convolutional neural networks.

INTRODUCTION: Chest X-rays (CXR) with under-exposure increase image noise and this may affect convolutional neural network (CNN) performance. This study aimed to train and validate CNNs for classifying pneumonia on CXR as normal or pneumonia acquired at different image noise levels. METHODS: The study used the curated and publicly available "Chest X-Ray Pneumonia" dataset of 5856 AP CXR classified into 1583 normal, 4273 viral and bacterial pneumonia cases. Gaussian noise with zero mean was added to the images, at 5 image noise variance levels, corresponding to decreasing exposure. Each noise-level dataset was split into 80% for training, 10% for validation, and 10% for test data and then classified using custom trained sequential CNN architecture. Six classification tasks were developed for five Gaussian noise levels and the original dataset. Sensitivity, specificity, predictive values and accuracy were used as evaluation performance metrics. RESULTS: CNN evaluation on the different datasets revealed no performance drop from the original dataset to the five datasets with different noise levels. Sensitivity, specificity and accuracy for the normal datasets were 98.7%, 76.1% and 90.2%. For the five Gaussian noise levels the sensitivity, specificity and accuracy ranged from 96.9% to 98.2%, 94.4%-98.7% and 96.8%-97.6%, respectively. A heat map was used for visual explanation of the CNNs. CONCLUSION: The CNNs sensitivity maintained, and the specificity increased in distinguishing between normal and pneumonia CXR with the introduction of image noise. IMPLICATIONS FOR PRACTICE: No performance drops of CNNs in distinguishing cases with and without pneumonia CXR with different Gaussian noise levels was observed. This has potential for decreasing radiation dose to patients or maintaining exposure parameters for patients that require additional radiographs.

Comparing Radiomics features of tumour and healthy liver tissue in a limited CT dataset: A machine learning study.

INTRODUCTION: Liver cancer lesions on Computed Tomography (CT) withholds a great amount of data, which is not visible to the radiologists and radiographer. Radiomics features can be extracted from the lesions and used to train Machine Learning (ML) algorithms to predict between tumour and liver tissue. The purpose of this study was to investigate and classify Radiomics features extracted from liver tumours and normal liver tissue in a limited CT dataset. METHODS: The Liver Tumour Segmentation Benchmark (LiTS) dataset consisting of 131 CT scans of the liver with segmentations of tumour tissue and healthy liver was used to extract Radiomic features. Extracted Radiomic features included size, shape, and location extracted with morphological and statistical techniques according to the International Symposium on Biomedical Imaging manual. Relevant features was selected with chi2 correlation and principal component analysis (PCA) with tumour and healthy liver tissue as outcome according to a consensus between three experienced radiologists. Logistic regression, random forest and support vector machine was used to train and validate the dataset with a 10-fold cross-validation method and the Grid Search as hyper-parameter tuning. Performance was evaluated with sensitivity, specificity and accuracy. RESULTS: The performance of the ML algorithms achieved sensitivities, specificities and accuracy ranging from 96.30% (95% CI: 81.03%-99.91%) to 100.00% (95% CI: 86.77%-100.00%), 91.30% (95% CI: 71.96%-98.93%) to 100.00% (95% CI: 83.89%-100.00%)and 94.00% (95% CI: 83.45%-98.75%) to 100.00% (95% CI: 92.45%-100.00%), respectively. CONCLUSION: ML algorithms classifies Radiomics features extracted from healthy liver and tumour tissue with perfect accuracy. The Radiomics signature allows for a prognostic biomarker for hepatic tumour screening on liver CT. IMPLICATIONS FOR PRACTICE: Differentiation between tumour and liver tissue with Radiomics ML algorithms have the potential to increase the diagnostic accuracy, assist in the decision-making of supplementary multiphasic enhanced medical imaging, as well as for developing novel prognostic biomarkers for liver cancer patients.

Detection of metallic objects on digital radiographs with convolutional neural networks: A MRI screening tool.

INTRODUCTION: Screening for metallic implants and foreign bodies before magnetic resonance imaging (MRI) examinations, are crucial for patient safety. History of health are supplied by the patient, a family member, screening of electronic health records or the picture and archive systems (PACS). PACS securely store and transmits digital radiographs (DR) and related reports with patient information. Convolutional neural networks (CNN) can be used to detect metallic objects in DRs stored in PACS. This study evaluates the accuracy of CNNs in the detection of metallic objects on DRs as an MRI screening tool. METHODS: The musculoskeletal radiographs (MURA) dataset consisting of 14.863 upper extremity studies were stratified into datasets with and without metal. For each anatomical region: Elbow, finger, hand, humerus, forearm, shoulder and wrist we trained and validated CNN algorithms to classify radiographs with and without metal. Algorithm performance was evaluated with area under the receiver-operating curve (AUC), sensitivity, specificity, predictive values and accuracies compared with a reference standard of manually labelling. RESULTS: Sensitivities, specificities and area under the ROC-curves (AUC) for the six anatomic regions ranged from 85.33% (95% CI: 78.64%-90.57%) to 100.00% (95% CI: 98.16%-100.00%), 75.44% (95% CI: 62.24%-85.87%) to 93.57% (95% CI: 88.78%-96.75%) and 0.95 to 0.99, respectively. CONCLUSION: CNN algorithms classify DRs with metallic objects for six different anatomic regions with near-perfect accuracy. The rapid and iterative capability of the algorithms allows for scalable expansion and as a substitute MRI screening tool for metallic objects. IMPLICATIONS FOR PRACTICE: All CNNs would be able to assist in metal detection of digital radiographs prior to MRI, an substantially decrease screening time.