Recommendations for the creation of benchmark datasets for reproducible artificial intelligence in radiology.

Various healthcare domains have witnessed successful preliminary implementation of artificial intelligence (AI) solutions, including radiology, though limited generalizability hinders their widespread adoption. Currently, most research groups and industry have limited access to the data needed for external validation studies. The creation and accessibility of benchmark datasets to validate such solutions represents a critical step towards generalizability, for which an array of aspects ranging from preprocessing to regulatory issues and biostatistical principles come into play. In this article, the authors provide recommendations for the creation of benchmark datasets in radiology, explain current limitations in this realm, and explore potential new approaches. CLINICAL RELEVANCE STATEMENT: Benchmark datasets, facilitating validation of AI software performance can contribute to the adoption of AI in clinical practice. KEY POINTS: Benchmark datasets are essential for the validation of AI software performance. Factors like image quality and representativeness of cases should be considered. Benchmark datasets can help adoption by increasing the trustworthiness and robustness of AI.

Research training during radiology residency: findings from the ESR Radiology Trainee Forum survey.

OBJECTIVES: To elucidate the research training exposure of radiology residents across ESR country members. METHODS: A 30-question survey was constructed by the Radiology Trainee Forum and was distributed among residents and subspecialty fellows of countries members of the ESR. The survey examined the training environment, the status of research training and publications among trainees, the conditions under which research was conducted, and the exposure to activities such as grant proposal preparation and manuscript reviewing. Descriptive statistics and the chi-square test were used to assess the responses to survey questions and evaluate factors related to these responses. RESULTS: A total of 159 participants from 29 countries provided fully completed questionnaires. Only 12/159 trainees already had a PhD degree and nearly half had never published a PubMed-indexed manuscript (76/159, 47.8%). Among those who published their papers during radiology training, most did so in the first or second year of residency (n = 26 and n = 20 participants, respectively). Most participants (79%) did not receive further statistical training during residency, fifty-five out of 159 (34.59%) respondents never had any guidance/training on how to read a paper and 58 out of 159 (36.48%) had never been encouraged to participate in any research. Most of them had worked after hours to carry out research at least a few times (47/159, 29.56%) or always (82/159, 51.57%). CONCLUSION: Analysis of research training among radiology trainees was performed. Areas for improvement were identified that can prompt changes in training curricula to prepare a highly competent European workforce. CRITICAL RELEVANCE STATEMENT: This survey has identified deficits in research training of radiology residents across countries members of ESR, pinpointing areas for improvement to fortify the future of radiology in Europe. KEY POINTS: Research exposure and training of radiology residents varies across countries and members of ESR. Radiology residents largely lack systematic research training, dedicated research time, and guidance. Areas for improvement in research training of radiology residents have been identified, aiding the fortification of radiology research across Europe.

Impact of AI on radiology: a EuroAIM/EuSoMII 2024 survey among members of the European Society of Radiology.

In order to assess the perceptions and expectations of the radiology staff about artificial intelligence (AI), we conducted an online survey among ESR members (January-March 2024). It was designed considering that conducted in 2018, updated according to recent advancements and emerging topics, consisting of seven questions regarding demographics and professional background and 28 AI questions. Of 28,000 members contacted, 572 (2%) completed the survey. AI impact was predominantly expected on breast and oncologic imaging, primarily involving CT, mammography, and MRI, and in the detection of abnormalities in asymptomatic subjects. About half of responders did not foresee an impact of AI on job opportunities. For 273/572 respondents (48%), AI-only reports would not be accepted by patients; and 242/572 respondents (42%) think that the use of AI systems will not change the relationship between the radiological team and the patient. According to 255/572 respondents (45%), radiologists will take responsibility for any AI output that may influence clinical decision-making. Of 572 respondents, 274 (48%) are currently using AI, 153 (27%) are not, and 145 (25%) are planning to do so. In conclusion, ESR members declare familiarity with AI technologies, as well as recognition of their potential benefits and challenges. Compared to the 2018 survey, the perception of AI's impact on job opportunities is in general slightly less optimistic (more positive from AI users/researchers), while the radiologist's responsibility for AI outputs is confirmed. The use of large language models is declared not only limited to research, highlighting the need for education in AI and its regulations. CRITICAL RELEVANCE STATEMENT: This study critically evaluates the current impact of AI on radiology, revealing significant usage patterns and clinical implications, thereby guiding future integration strategies to enhance efficiency and patient care in clinical radiology. KEY POINTS: The survey examines ESR member's views about the impact of AI on radiology practice. AI use is relevant in CT and MRI, with varying impacts on job roles. AI tools enhance clinical efficiency but require radiologist oversight for patient acceptance.

Musculoskeletal Infection: The Great Mimickers on Imaging.

A series of conditions can mimic musculoskeletal infections on imaging, complicating their diagnosis and affecting the treatment. Depending on the anatomical location, different conditions can manifest with clinical and imaging findings that mimic infections. Herein we present a wide spectrum of the musculoskeletal disorders of the axial skeleton, long bones, peripheral joints, and soft tissue that may manifest as infectious processes, and we focus on the potential mimics of osteomyelitis, septic arthritis, and infectious spondylodiscitis that are common in clinical practice. We present the typical imaging characteristics of each musculoskeletal infection, followed by mimicking conditions.

Diagnostic performance of radiomics in prediction of Ki-67 index status in non-small cell lung cancer: A systematic review and meta-analysis.

BACKGROUND: Lung cancer's high prevalence and invasiveness make it a major global health concern. The Ki-67 index, which indicates cellular proliferation, is crucial for assessing lung cancer aggressiveness. Radiomics, which extracts quantifiable features from medical images using algorithms, may provide insights into tumor behavior. This systematic review and meta-analysis evaluate the effectiveness of radiomics in predicting Ki-67 status in Non-Small Cell Lung Cancer (NSCLC) using CT scans. METHODS AND MATERIALS: A comprehensive search was conducted in PubMed/MEDLINE, Embase, Scopus, and Web of Science databases from inception until April 19, 2024. Original studies discussing the performance of CT-based radiomics for predicting Ki-67 status in NSCLC cohorts were included. The quality assessment involved quality assessment of diagnostic accuracy studies (QUADAS-2), radiomics quality score (RQS) and METhodological RadiomICs Score (METRICS). Quantitative meta-analysis, using R, assessed pooled diagnostic odds ratio, sensitivity, and specificity in NSCLC cohorts. RESULTS: We identified 10 studies that met the inclusion criteria, involving 2279 participants, with 9 of these studies included in quantitative meta-analysis. The pooled sensitivity and specificity of radiomics-based models for predicting Ki-67 status in NSCLC were 0.783 (95 % CI: 0.732 - 0.827) and 0.796 (95 % CI: 0.707 - 0.864) in training cohorts, and 0.803 (95 % CI: 0.744 - 0.851) and 0.696 (95 % CI: 0.613 - 0.768) in validation cohorts. It was identified in subgroup analysis that utilizing ITK-SNAP as a segmentation software contributed to a significantly higher pooled sensitivity. CONCLUSION: This meta-analysis indicates promising diagnostic accuracy of radiomics in predicting Ki-67 in NSCLC.

Prognostic value of CT scan-based radiomics in intracerebral hemorrhage patients: A systematic review and meta-analysis.

OBJECTIVES: We conducted a systematic review and meta-analysis of current publications on the potential role of non-contrast-enhanced computed tomography (NCCT) radiomics as a prognostic indicator in patients with intracerebral hemorrhage (ICH). METHODS: We systematically searched PubMed, EMBASE, and the Web of Science from inception until January 8, 2024. Studies with NCCT-based radiomics features for predicting the prognostic outcomes of ICH patients were included. We calculated the pooled sensitivity, specificity, diagnostic odds ratio (DOR), and area under curve (AUC) values. The radiomics quality score (RQS), METhodological RadiomICs Score (METRICS), and the quality assessment of diagnostic accuracy studies (QUADAS-2) were used for quality assessment. RESULTS: Twenty-two studies were included. The pooled sensitivity, specificity, DOR, and AUC of radiomics models were 0.73, 0.78, 10.03, and 0.83, respectively, while on the combined radiomics models with other non-radiomics features were 0.80, 0.80, 16.28, and 0.86. Subgroup analysis showed that studies with the following covariates have a higher accuracy: single center, modified Rankin Scale (mRS) criteria for the ICH outcomes assessment, following patients for evaluation of ICH outcomes for more than a month, automatic segmentation, capturing the radiomics feature from the only intra-hematomal region, using PyRadiomic tool for features extraction, and using non-logistic regression for modeling. The quality of literature using QUADAS-2 and METRICS tools was good and was under-average using the RQS tool. No publication bias was detected. CONCLUSIONS: Radiomics features showed moderate to high accuracy for predicting ICH prognostic outcomes. Although the QUADAS-2 and METRICS assessments indicated good quality, the radiomics pipeline quality was under-average. CLINICAL RELEVANCE: NCCT-based radiomics features can provide information about the prognostic outcomes of ICH patients after patient admission. This study exploits the value of current evidence on NCCT-based radiomics methodology in the prediction of ICH prognosis.

Bias in artificial intelligence for medical imaging: fundamentals, detection, avoidance, mitigation, challenges, ethics, and prospects.

Although artificial intelligence (AI) methods hold promise for medical imaging-based prediction tasks, their integration into medical practice may present a double-edged sword due to bias (i.e., systematic errors). AI algorithms have the potential to mitigate cognitive biases in human interpretation, but extensive research has highlighted the tendency of AI systems to internalize biases within their model. This fact, whether intentional or not, may ultimately lead to unintentional consequences in the clinical setting, potentially compromising patient outcomes. This concern is particularly important in medical imaging, where AI has been more progressively and widely embraced than any other medical field. A comprehensive understanding of bias at each stage of the AI pipeline is therefore essential to contribute to developing AI solutions that are not only less biased but also widely applicable. This international collaborative review effort aims to increase awareness within the medical imaging community about the importance of proactively identifying and addressing AI bias to prevent its negative consequences from being realized later. The authors began with the fundamentals of bias by explaining its different definitions and delineating various potential sources. Strategies for detecting and identifying bias were then outlined, followed by a review of techniques for its avoidance and mitigation. Moreover, ethical dimensions, challenges encountered, and prospects were discussed.

Prediction of Anastomotic Leakage in Esophageal Cancer Surgery: A Multimodal Machine Learning Model Integrating Imaging and Clinical Data.

RATIONALE AND OBJECTIVES: Surgery in combination with chemo/radiotherapy is the standard treatment for locally advanced esophageal cancer. Even after the introduction of minimally invasive techniques, esophagectomy carries significant morbidity and mortality. One of the most common and feared complications of esophagectomy is anastomotic leakage (AL). Our work aimed to develop a multimodal machine-learning model combining CT-derived and clinical data for predicting AL following esophagectomy for esophageal cancer. MATERIAL AND METHODS: A total of 471 patients were prospectively included (Jan 2010-Dec 2022). Preoperative computed tomography (CT) was used to evaluate celia trunk stenosis and vessel calcification. Clinical variables, including demographics, disease stage, operation details, postoperative CRP, and stage, were combined with CT data to build a model for AL prediction. Data was split into 80%:20% for training and testing, and an XGBoost model was developed with 10-fold cross-validation and early stopping. ROC curves and respective areas under the curve (AUC), sensitivity, specificity, PPV, NPV, and F1-scores were calculated. RESULTS: A total of 117 patients (24.8%) exhibited post-operative AL. The XGboost model achieved an AUC of 79.2% (95%CI 69%-89.4%) with a specificity of 77.46%, a sensitivity of 65.22%, PPV of 48.39%, NPV of 87.3%, and F1-score of 56%. Shapley Additive exPlanation analysis showed the effect of individual variables on the result of the model. Decision curve analysis showed that the model was particularly beneficial for threshold probabilities between 15% and 48%. CONCLUSION: A clinically relevant multimodal model can predict AL, which is especially valuable in cases with low clinical probability of AL.

Utility of Dual-Energy Computed Tomography in lesion characterization and treatment planning for peripheral Chronic Total Occlusions: A comprehensive analysis of crossing difficulty.

PURPOSE: To investigate whether Dual-Energy Computed Tomography (DECT) could be useful in the lesion characterization and endovascular treatment planning of symptomatic patients with peripheral arterial disease (PAD) due to Chronic Total Occlusions (CTO). MATERIALS AND METHODS: Between 2018 and 2022, 60 symptomatic patients (52 male, age 71 years) with peripheral arterial CTO underwent DECT angiography before percutaneous endovascular treatment. Patients were classified, according to guidewire crossing difficulty into four categories, which were subsequently correlated with DECT values, including Dual Energy Index (DEI) and Effective Z (Zeff). DECT values were also corelated with crossing time. The crossing difficulty was further correlated with the Trans-Atlantic Inter-Society Consensus Document (TASC II) classification. RESULTS: Technical success, defined as perceived antegrade true lumen or subintimal crossing, was achieved in 76.7 %. Among the cases, 20 were deemed easy, 14 moderate, 12 hard and 14 were failed attempts. Statistical analysis revealed a significant correlation between DEI, Zeff values, and the crossing difficulty categories (p < 0.001). Additionally, there was also a correlation between crossing time and DECT values. However, no significant correlation was recorded between difficulty categories and TASC II classification. CONCLUSION: Pre-procedural DECT angiography provides valuable information for patient selection and planning of the revascularization strategy. Moreover, it is helpful in the selection of the appropriate PTA materials, based on the lesion characteristics. Further research should be invested in this important field, to determine the optimal treatment approach in patients suffering from PAD due to CTOs.

Performance of thoracic ultrasonography compared with chest radiography for the detection of rib fractures using computed tomography as a reference standard.

OBJECTIVE: Although there is growing evidence that ultrasonography is superior to X-ray for rib fractures' detection, X-ray is still indicated as the most appropriate method. This has partially been attributed to a lack of studies using an appropriate reference modality. We aimed to compare the diagnostic accuracy of ultrasonography and X-ray in the detection of rib fractures, considering CT as the reference standard. MATERIALS AND METHODS: Within a 2.5-year period, all consecutive patients with clinically suspected rib fracture(s) following blunt chest trauma and available posteroanterior/anteroposterior X-ray and thoracic CT were prospectively studied and planned to undergo thoracic ultrasonography, by a single operator. All imaging examinations were evaluated for cortical rib fracture(s), and their location was recorded. The cartilaginous rib portions were not assessed. CTs and X-rays were evaluated retrospectively. Concomitant thoracic/extra-thoracic injuries were assessed on CT. Comparisons were performed with the Mann-Whitney U test and Fisher's exact test. RESULTS: Fifty-nine patients (32 males, 27 females; mean age, 53.1 ± 16.6 years) were included. CT, ultrasonography, and X-ray (40 posteroanterior/19 anteroposterior views) diagnosed 136/122/42 rib fractures in 56/54/27 patients, respectively. Ultrasonography and X-ray had sensitivity of 100%/40% and specificity of 89.7%/30.9% for rib fractures' detection. Ultrasound accuracy was 94.9% compared to 35.4% for X-rays (P < .001) in detecting individual rib fractures. Most fractures involved the 4th-9th ribs. Upper rib fractures were most commonly overlooked on ultrasonography. Thoracic cage/spine fractures and haemothorax represented the most common concomitant injuries. CONCLUSION: Ultrasonography appeared to be superior to X-ray for the detection of rib fractures with regard to a reference CT.

How young radiologists use contrast media and manage adverse reactions: an international survey.

OBJECTIVES: To collect real-world data about the knowledge and self-perception of young radiologists concerning the use of contrast media (CM) and the management of adverse drug reactions (ADR). METHODS: A survey (29 questions) was distributed to residents and board-certified radiologists younger than 40 years to investigate the current international situation in young radiology community regarding CM and ADRs. Descriptive statistics analysis was performed. RESULTS: Out of 454 respondents from 48 countries (mean age: 31.7 ± 4 years, range 25-39), 271 (59.7%) were radiology residents and 183 (40.3%) were board-certified radiologists. The majority (349, 76.5%) felt they were adequately informed regarding the use of CM. However, only 141 (31.1%) received specific training on the use of CM and 82 (18.1%) about management ADR during their residency. Although 266 (58.6%) knew safety protocols for handling ADR, 69.6% (316) lacked confidence in their ability to manage CM-induced ADRs and 95.8% (435) expressed a desire to enhance their understanding of CM use and handling of CM-induced ADRs. Nearly 300 respondents (297; 65.4%) were aware of the benefits of contrast-enhanced ultrasound, but 249 (54.8%) of participants did not perform it. The preferred CM injection strategy in CT parenchymal examination and CT angiography examination was based on patient's lean body weight in 318 (70.0%) and 160 (35.2%), a predeterminate fixed amount in 79 (17.4%) and 116 (25.6%), iodine delivery rate in 26 (5.7%) and 122 (26.9%), and scan time in 31 (6.8%) and 56 (12.3%), respectively. CONCLUSION: Training in CM use and management ADR should be implemented in the training of radiology residents. CRITICAL RELEVANCE STATEMENT: We highlight the need for improvement in the education of young radiologists regarding contrast media; more attention from residency programs and scientific societies should be focused on training about contrast media use and the management of adverse drug reactions. KEY POINTS: • This survey investigated training of young radiologists about use of contrast media and management adverse reactions. • Most young radiologists claimed they did not receive dedicated training. • An extreme heterogeneity of responses was observed about contrast media indications/contraindications and injection strategy.

Developing a Radiomics Atlas Dataset of normal Abdominal and Pelvic computed Tomography (RADAPT).

Atlases of normal genomics, transcriptomics, proteomics, and metabolomics have been published in an attempt to understand the biological phenotype in health and disease and to set the basis of comprehensive comparative omics studies. No such atlas exists for radiomics data. The purpose of this study was to systematically create a radiomics dataset of normal abdominal and pelvic radiomics that can be used for model development and validation. Young adults without any previously known disease, aged > 17 and ≤ 36 years old, were retrospectively included. All patients had undergone CT scanning for emergency indications. In case abnormal findings were identified, the relevant anatomical structures were excluded. Deep learning was used to automatically segment the majority of visible anatomical structures with the TotalSegmentator model as applied in 3DSlicer. Radiomics features including first order, texture, wavelet, and Laplacian of Gaussian transformed features were extracted with PyRadiomics. A Github repository was created to host the resulting dataset. Radiomics data were extracted from a total of 531 patients with a mean age of 26.8 ± 5.19 years, including 250 female and 281 male patients. A maximum of 53 anatomical structures were segmented and used for subsequent radiomics data extraction. Radiomics features were derived from a total of 526 non-contrast and 400 contrast-enhanced (portal venous) series. The dataset is publicly available for model development and validation purposes.

Perspectives for the Use of Umbilical Cord Blood in Transplantation and Beyond: Initiatives for an Advanced and Sustainable Public Banking Program in Greece.

The umbilical cord blood (UCB) donated in public UCB banks is a source of hematopoietic stem cells (HSC) alternative to bone marrow for allogeneic HSC transplantation (HSCT). However, the high rejection rate of the donated units due to the strict acceptance criteria and the wide application of the haploidentical HSCT have resulted in significant limitation of the use of UCB and difficulties in the economic sustainability of the public UCB banks. There is an ongoing effort within the UCB community to optimize the use of UCB in the field of HSCT and a parallel interest in exploring the use of UCB for applications beyond HSCT i.e., in the fields of cell therapy, regenerative medicine and specialized transfusion medicine. In this report, we describe the mode of operation of the three public UCB banks in Greece as an example of an orchestrated effort to develop a viable UCB banking system by (a) prioritizing the enrichment of the national inventory by high-quality UCB units from populations with rare human leukocyte antigens (HLA), and (b) deploying novel sustainable applications of UCB beyond HSCT, through national and international collaborations. The Greek paradigm of the public UCB network may become an example for countries, particularly with high HLA heterogeneity, with public UCB banks facing sustainability difficulties and adds value to the international efforts aiming to sustainably expand the public UCB banking system.

Convolutional neural networks for the differentiation between benign and malignant renal tumors with a multicenter international computed tomography dataset.

OBJECTIVES: To use convolutional neural networks (CNNs) for the differentiation between benign and malignant renal tumors using contrast-enhanced CT images of a multi-institutional, multi-vendor, and multicenter CT dataset. METHODS: A total of 264 histologically confirmed renal tumors were included, from US and Swedish centers. Images were augmented and divided randomly 70%:30% for algorithm training and testing. Three CNNs (InceptionV3, Inception-ResNetV2, VGG-16) were pretrained with transfer learning and fine-tuned with our dataset to distinguish between malignant and benign tumors. The ensemble consensus decision of the three networks was also recorded. Performance of each network was assessed with receiver operating characteristics (ROC) curves and their area under the curve (AUC-ROC). Saliency maps were created to demonstrate the attention of the highest performing CNN. RESULTS: Inception-ResNetV2 achieved the highest AUC of 0.918 (95% CI 0.873-0.963), whereas VGG-16 achieved an AUC of 0.813 (95% CI 0.752-0.874). InceptionV3 and ensemble achieved the same performance with an AUC of 0.894 (95% CI 0.844-0.943). Saliency maps indicated that Inception-ResNetV2 decisions are based on the characteristics of the tumor while in most tumors considering the characteristics of the interface between the tumor and the surrounding renal parenchyma. CONCLUSION: Deep learning based on a diverse multicenter international dataset can enable accurate differentiation between benign and malignant renal tumors. CRITICAL RELEVANCE STATEMENT: Convolutional neural networks trained on a diverse CT dataset can accurately differentiate between benign and malignant renal tumors. KEY POINTS: • Differentiation between benign and malignant tumors based on CT is extremely challenging. • Inception-ResNetV2 trained on a diverse dataset achieved excellent differentiation between tumor types. • Deep learning can be used to distinguish between benign and malignant renal tumors.

Generative adversarial networks for spine imaging: A critical review of current applications.

PURPOSE: In recent years, the field of medical imaging has witnessed remarkable advancements, with innovative technologies which revolutionized the visualization and analysis of the human spine. Among the groundbreaking developments in medical imaging, Generative Adversarial Networks (GANs) have emerged as a transformative tool, offering unprecedented possibilities in enhancing spinal imaging techniques and diagnostic outcomes. This review paper aims to provide a comprehensive overview of the use of GANs in spinal imaging, and to emphasize their potential to improve the diagnosis and treatment of spine-related disorders. A specific review focusing on Generative Adversarial Networks (GANs) in the context of medical spine imaging is needed to provide a comprehensive and specialized analysis of the unique challenges, applications, and advancements within this specific domain, which might not be fully addressed in broader reviews covering GANs in general medical imaging. Such a review can offer insights into the tailored solutions and innovations that GANs bring to the field of spinal medical imaging. METHODS: An extensive literature search from 2017 until July 2023, was conducted using the most important search engines and identified studies that used GANs in spinal imaging. RESULTS: The implementations include generating fat suppressed T2-weighted (fsT2W) images from T1 and T2-weighted sequences, to reduce scan time. The generated images had a significantly better image quality than true fsT2W images and could improve diagnostic accuracy for certain pathologies. GANs were also utilized in generating virtual thin-slice images of intervertebral spaces, creating digital twins of human vertebrae, and predicting fracture response. Lastly, they could be applied to convert CT to MRI images, with the potential to generate near-MR images from CT without MRI. CONCLUSIONS: GANs have promising applications in personalized medicine, image augmentation, and improved diagnostic accuracy. However, limitations such as small databases and misalignment in CT-MRI pairs, must be considered.