The objective is to assess the performance of seven semiautomatic and two fully automatic segmentation methods on [F-18]FDG PET/CT lymphoma images and evaluate their influence on tumor... Show moreThe objective is to assess the performance of seven semiautomatic and two fully automatic segmentation methods on [F-18]FDG PET/CT lymphoma images and evaluate their influence on tumor quantification. All lymphoma lesions identified in 65 whole-body [F-18]FDG PET/CT staging images were segmented by two experienced observers using manual and semiautomatic methods. Semiautomatic segmentation using absolute and relative thresholds, k-means and Bayesian clustering, and a self-adaptive configuration (SAC) of k-means and Bayesian was applied. Three state-of-the-art deep learning-based segmentations methods using a 3D U-Net architecture were also applied. One was semiautomatic and two were fully automatic, of which one is publicly available. Dice coefficient (DC) measured segmentation overlap, considering manual segmentation the ground truth. Lymphoma lesions were characterized by 31 features. Intraclass correlation coefficient (ICC) assessed features agreement between different segmentation methods. Nine hundred twenty [F-18]FDG-avid lesions were identified. The SAC Bayesian method achieved the highest median intra-observer DC (0.87). Inter-observers' DC was higher for SAC Bayesian than manual segmentation (0.94 vs 0.84, p < 0.001). Semiautomatic deep learning-based median DC was promising (0.83 (Obs1), 0.79 (Obs2)). Threshold-based methods and publicly available 3D U-Net gave poorer results (0.56 <= DC <= 0.68). Maximum, mean, and peak standardized uptake values, metabolic tumor volume, and total lesion glycolysis showed excellent agreement (ICC >= 0.92) between manual and SAC Bayesian segmentation methods. The SAC Bayesian classifier is more reproducible and produces similar lesion features compared to manual segmentation, giving the best concordant results of all other methods. Deep learning-based segmentation can achieve overall good segmentation results but failed in few patients impacting patients' clinical evaluation. Show less
BackgroundMeasurement of peak velocities is important in the evaluation of heart failure. This study compared the performance of automated 4D flow cardiac MRI (CMR) with traditional transthoracic... Show moreBackgroundMeasurement of peak velocities is important in the evaluation of heart failure. This study compared the performance of automated 4D flow cardiac MRI (CMR) with traditional transthoracic Doppler echocardiography (TTE) for the measurement of mitral inflow peak diastolic velocities.MethodsPatients with Doppler echocardiography and 4D flow cardiac magnetic resonance data were included retrospectively. An established automated technique was used to segment the left ventricular transvalvular flow using short-axis cine stack of images. Peak mitral E-wave and peak mitral A-wave velocities were automatically derived using in-plane velocity maps of transvalvular flow. Additionally, we checked the agreement between peak mitral E-wave velocity derived by 4D flow CMR and Doppler echocardiography in patients with sinus rhythm and atrial fibrillation (AF) separately.ResultsForty-eight patients were included (median age 69 years, IQR 63 to 76; 46% female). Data were split into three groups according to heart rhythm. The median peak E-wave mitral inflow velocity by automated 4D flow CMR was comparable with Doppler echocardiography in all patients (0.90 +/- 0.43 m/s vs 0.94 +/- 0.48 m/s, P = 0.132), sinus rhythm-only group (0.88 +/- 0.35 m/s vs 0.86 +/- 0.38 m/s, P = 0.54) and in AF-only group (1.33 +/- 0.56 m/s vs 1.18 +/- 0.47 m/s, P = 0.06). Peak A-wave mitral inflow velocity results had no significant difference between Doppler TTE and automated 4D flow CMR (0.81 +/- 0.44 m/s vs 0.81 +/- 0.53 m/s, P = 0.09) in all patients and sinus rhythm-only groups. Automated 4D flow CMR showed a significant correlation with TTE for measurement of peak E-wave in all patients group (r = 0.73, P < 0.001) and peak A-wave velocities (r = 0.88, P < 0.001). Moreover, there was a significant correlation between automated 4D flow CMR and TTE for peak-E wave velocity in sinus rhythm-only patients (r = 0.68, P < 0.001) and AF-only patients (r = 0.81, P = 0.014). Excellent intra-and inter-observer variability was demonstrated for both parameters.ConclusionAutomated dynamic peak mitral inflow diastolic velocity tracing using 4D flow CMR is comparable to Doppler echocardiography and has excellent repeatability for clinical use. However, 4D flow CMR can potentially underestimate peak velocity in patients with AF. Show less
Unterrainer, M.; Deroose, C.M.; Herrmann, K.; Moehler, M.; Blomqvist, L.; Cannella, R.; ... ; European Soc Gastrointestinal Abdominal Radiology (ESGAR) 2022
Background: Treatment monitoring in metastatic colorectal cancer (mCRC) relies on imaging to evaluate the tumour burden. Response Evaluation Criteria in Solid Tumors provide a framework on... Show moreBackground: Treatment monitoring in metastatic colorectal cancer (mCRC) relies on imaging to evaluate the tumour burden. Response Evaluation Criteria in Solid Tumors provide a framework on reporting and interpretation of imaging findings yet offer no guidance on a standardised imaging protocol tailored to patients with mCRC. Imaging protocol hetero-geneity remains a challenge for the reproducibility of conventional imaging end-points and is an obstacle for research on novel imaging end-points. Patients and methods: Acknowledging the recently highlighted potential of radiomics and arti-ficial intelligence tools as decision support for patient care in mCRC, a multidisciplinary, international and expert panel of imaging specialists was formed to find consensus on mCRC imaging protocols using the Delphi method. Results: Under the guidance of the European Organisation for Research and Treatment of Cancer (EORTC) Imaging and Gastrointestinal Tract Cancer Groups, the European Society of Oncologic Imaging (ESOI) and the European Society of Gastrointestinal and Abdominal Radiology (ESGAR), the EORTC-ESOI-ESGAR core imaging protocol was identified. Conclusion: This consensus protocol attempts to promote standardisation and to diminish variations in patient preparation, scan acquisition and scan reconstruction. We anticipate that this standardisation will increase reproducibility of radiomics and artificial intelligence studies and serve as a catalyst for future research on imaging end-points. For ongoing and future mCRC trials, we encourage principal investigators to support the dissemination of these im-aging standards across recruiting centres. (c) 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Show less
The societal burden of spinal conditions is vast and continues to grow with the in- creasing prevalence of patients with spinal degenerative disease, spinal metasta- ses, and spinal infections.... Show moreThe societal burden of spinal conditions is vast and continues to grow with the in- creasing prevalence of patients with spinal degenerative disease, spinal metasta- ses, and spinal infections. Recent application of artificial intelligence in healthcare have shown great promise and similar extensions in spine surgery may improve decision-making. The purpose of this thesis was to examine the utility of predictive analytics and natural language processing in spine surgery. Show less
Background: Timely identification of deteriorating COVID-19 patients is needed to guide changes in clinical management and admission to intensive care units (ICUs). There is significant concern... Show moreBackground: Timely identification of deteriorating COVID-19 patients is needed to guide changes in clinical management and admission to intensive care units (ICUs). There is significant concern that widely used Early warning scores (EWSs) underestimate illness severity in COVID-19 patients and therefore, we developed an early warning model specifically for COVID-19 patients. Methods: We retrospectively collected electronic medical record data to extract predictors and used these to fit a random forest model. To simulate the situation in which the model would have been developed after the first and implemented during the second COVID-19 `wave' in the Netherlands, we performed a temporal validation by splitting all included patients into groups admitted before and after August 1, 2020. Furthermore, we propose a method for dynamic model updating to retain model performance over time. We evaluated model discrimination and calibration, performed a decision curve analysis, and quantified the importance of predictors using SHapley Additive exPlanations values. Results: We included 3514 COVID-19 patient admissions from six Dutch hospitals between February 2020 and May 2021, and included a total of 18 predictors for model fitting. The model showed a higher discriminative performance in terms of partial area under the receiver operating characteristic curve (0.82 [0.80-0.84]) compared to the National early warning score (0.72 [0.69-0.74]) and the Modified early warning score (0.67 [0.65-0.69]), a greater net benefit over a range of clinically relevant model thresholds, and relatively good calibration (intercept = 0.03 [- 0.09 to 0.14], slope = 0.79 [0.73-0.86]). Conclusions: This study shows the potential benefit of moving from early warning models for the general inpatient population to models for specific patient groups. Further (independent) validation of the model is needed. Show less
Flexible high-definition white-light endoscopy is the current gold standard in screening for cancer and its precursor lesions in the gastrointestinal tract. However, miss rates are high, especially... Show moreFlexible high-definition white-light endoscopy is the current gold standard in screening for cancer and its precursor lesions in the gastrointestinal tract. However, miss rates are high, especially in populations at high risk for developing gastrointestinal cancer (e.g., inflammatory bowel disease, Lynch syndrome, or Barrett's esophagus) where lesions tend to be flat and subtle. Fluorescence molecular endoscopy (FME) enables intraluminal visualization of (pre)malignant lesions based on specific biomolecular features rather than morphology by using fluorescently labeled molecular probes that bind to specific molecular targets. This strategy has the potential to serve as a valuable tool for the clinician to improve endoscopic lesion detection and real-time clinical decision-making. This narrative review presents an overview of recent advances in FME, focusing on probe development, techniques, and clinical evidence. Future perspectives will also be addressed, such as the use of FME in patient stratification for targeted therapies and potential alliances with artificial intelligence. Key Messages center dot Fluorescence molecular endoscopy is a relatively new technology that enables safe and real-time endoscopic lesion visualization based on specific molecular features rather than on morphology, thereby adding a layer of information to endoscopy, like in PET-CT imaging. center dot Recently the transition from preclinical to clinical studies has been made, with promising results regarding enhancing detection of flat and subtle lesions in the colon and esophagus. However, clinical evidence needs to be strengthened by larger patient studies with stratified study designs. center dot In the future fluorescence molecular endoscopy could serve as a valuable tool in clinical workflows to improve detection in high-risk populations like patients with Barrett's esophagus, Lynch syndrome, and inflammatory bowel syndrome, where flat and subtle lesions tend to be malignant up to five times more often. center dot Fluorescence molecular endoscopy has the potential to assess therapy responsiveness in vivo for targeted therapies, thereby playing a role in personalizing medicine. center dot To further reduce high miss rates due to human and technical factors, joint application of artificial intelligence and fluorescence molecular endoscopy are likely to generate added value. Show less
BACKGROUND: Robotic neurosurgery may improve the accuracy, speed, and availability of stereotactic procedures. We recently developed a computer vision and artificial intelligence-driven frameless... Show moreBACKGROUND: Robotic neurosurgery may improve the accuracy, speed, and availability of stereotactic procedures. We recently developed a computer vision and artificial intelligence-driven frameless stereotaxy for nonimmobilized patients, creating an opportunity to develop accurate and rapidly deployable robots for bedside cranial intervention. OBJECTIVE: To validate a portable stereotactic surgical robot capable of frameless registration, real-time tracking, and accurate bedside catheter placement. METHODS: Four human cadavers were used to evaluate the robot's ability to maintain low surface registration and targeting error for 72 intracranial targets during head motion, ie, without rigid cranial fixation. Twenty-four intracranial catheters were placed robotically at predetermined targets. Placement accuracy was verified by computed tomography imaging. RESULTS: Robotic tracking of the moving cadaver heads occurred with a program runtime of 0.111 +/- 0.013 seconds, and the movement command latency was only 0.002 +/- 0.003 seconds. For surface error tracking, the robot sustained a 0.588 +/- 0.105 mm registration accuracy during dynamic head motions (velocity of 6.647 +/- 2.360 cm/s). For the 24 robotic-assisted intracranial catheter placements, the target registration error was 0.848 +/- 0.590 mm, providing a user error of 0.339 +/- 0.179 mm. CONCLUSION: Robotic-assisted stereotactic procedures on mobile subjects were feasible with this robot and computer vision image guidance technology. Frameless robotic neurosurgery potentiates surgery on nonimmobilized and awake patients both in the operating room and at the bedside. It can affect the field through improving the safety and ability to perform procedures such as ventriculostomy, stereo electroencephalography, biopsy, and potentially other novel procedures. If we envision catheter misplacement as a "never event," robotics can facilitate that reality. Show less
Purpose The tumor-stroma ratio (TSR) has repeatedly proven to be correlated with patient outcomes in breast cancer using large retrospective cohorts. However, studies validating the TSR often show... Show morePurpose The tumor-stroma ratio (TSR) has repeatedly proven to be correlated with patient outcomes in breast cancer using large retrospective cohorts. However, studies validating the TSR often show variability in methodology, thereby hampering comparisons and uniform outcomes. Method This paper provides a detailed description of a simple and uniform TSR scoring method using Hematoxylin and Eosin (H&E)-stained core biopsies and resection tissue, specifically focused on breast cancer. Possible histological challenges that can be encountered during scoring including suggestions to overcome them are reported. Moreover, the procedure for TSR estimation in lymph nodes, scoring on digital images and the automatic assessment of the TSR using artificial intelligence are described. Conclusion Digitized scoring of tumor biopsies and resection material offers interesting future perspectives to determine patient prognosis and response to therapy. The fact that the TSR method is relatively easy, quick, and cheap, offers great potential for its implementation in routine diagnostics, but this requires high quality validation studies. Show less
Background Right atrial (RA) area predicts mortality in patients with pulmonary hypertension, and is recommended by the European Society of Cardiology/European Respiratory Society pulmonary... Show moreBackground Right atrial (RA) area predicts mortality in patients with pulmonary hypertension, and is recommended by the European Society of Cardiology/European Respiratory Society pulmonary hypertension guidelines. The advent of deep learning may allow more reliable measurement of RA areas to improve clinical assessments. The aim of this study was to automate cardiovascular magnetic resonance (CMR) RA area measurements and evaluate the clinical utility by assessing repeatability, correlation with invasive haemodynamics and prognostic value. Methods A deep learning RA area CMR contouring model was trained in a multicentre cohort of 365 patients with pulmonary hypertension, left ventricular pathology and healthy subjects. Inter-study repeatability (intraclass correlation coefficient (ICC)) and agreement of contours (DICE similarity coefficient (DSC)) were assessed in a prospective cohort (n = 36). Clinical testing and mortality prediction was performed in n = 400 patients that were not used in the training nor prospective cohort, and the correlation of automatic and manual RA measurements with invasive haemodynamics assessed in n = 212/400. Radiologist quality control (QC) was performed in the ASPIRE registry, n = 3795 patients. The primary QC observer evaluated all the segmentations and recorded them as satisfactory, suboptimal or failure. A second QC observer analysed a random subcohort to assess QC agreement (n = 1018). Results All deep learning RA measurements showed higher interstudy repeatability (ICC 0.91 to 0.95) compared to manual RA measurements (1st observer ICC 0.82 to 0.88, 2nd observer ICC 0.88 to 0.91). DSC showed high agreement comparing automatic artificial intelligence and manual CMR readers. Maximal RA area mean and standard deviation (SD) DSC metric for observer 1 vs observer 2, automatic measurements vs observer 1 and automatic measurements vs observer 2 is 92.4 +/- 3.5 cm(2), 91.2 +/- 4.5 cm(2) and 93.2 +/- 3.2 cm(2), respectively. Minimal RA area mean and SD DSC metric for observer 1 vs observer 2, automatic measurements vs observer 1 and automatic measurements vs observer 2 was 89.8 +/- 3.9 cm(2), 87.0 +/- 5.8 cm(2) and 91.8 +/- 4.8 cm(2). Automatic RA area measurements all showed moderate correlation with invasive parameters (r = 0.45 to 0.66), manual (r = 0.36 to 0.57). Maximal RA area could accurately predict elevated mean RA pressure low and high-risk thresholds (area under the receiver operating characteristic curve artificial intelligence = 0.82/0.87 vs manual = 0.78/0.83), and predicted mortality similar to manual measurements, both p < 0.01. In the QC evaluation, artificial intelligence segmentations were suboptimal at 108/3795 and a low failure rate of 16/3795. In a subcohort (n = 1018), agreement by two QC observers was excellent, kappa 0.84. Conclusion Automatic artificial intelligence CMR derived RA size and function are accurate, have excellent repeatability, moderate associations with invasive haemodynamics and predict mortality. Show less
The aim of this thesis is to determine diagnostic performance of machine learning in differentiating between atypical cartilaginous tumor (ACT) and high-grade chondrosarcoma (CS) based on radiomic... Show moreThe aim of this thesis is to determine diagnostic performance of machine learning in differentiating between atypical cartilaginous tumor (ACT) and high-grade chondrosarcoma (CS) based on radiomic features derived from magnetic resonance imaging (MRI) and computed tomography (CT). In chapter 2, the concept of radiomics of musculoskeletal sarcomas is introduced and a systematic review on radiomic feature reproducibility and validation strategies is conducted. In chapter 3, a preliminary study is performed to investigate the performance of MRI radiomics-based machine learning in discriminating ACT from high-grade CS, using a single-center cohort, in comparison with an expert radiologist. In chapter 4, the influence of interobserver segmentation variability on the reproducibility of CT and MRI radiomic features of cartilaginous bone tumors is assessed. In chapter 5, the performance of CT radiomics-based machine learning in discriminating ACT from high-grade CS of long bones is determined and validated using independent data from a multicenter cohort, compared to an expert radiologist. In chapter 6, the performance of MRI radiomics-based machine learning in differentiating between ACT and grade II CS of long bones is determined and validated using independent data from a multicenter cohort, in comparison with an expert radiologist. Finally, in chapter 7, the main results and implications of this thesis are summarized and discussed. Show less
Despite improved surgical and adjuvant treatment options, malignant brain tumors remain non-curable to date. The thin line between treatment effectiveness and patient harms underpins the importance... Show moreDespite improved surgical and adjuvant treatment options, malignant brain tumors remain non-curable to date. The thin line between treatment effectiveness and patient harms underpins the importance of tailoring clinical management to the individual brain tumor patient. Over the past decades, the volume and complexity of clinically-derived patient data (i.e., imaging, genomics, free-text etc.) is increasing exponentially. Machine learning provides a vast range of algorithms that can learn from this data and guide clinical decision-making by providing accurate patient-level predictions. The current thesis describes several studies along the continuum of the machine learning spectrum as it applies to neurosurgical oncology. Part I investigates postoperative complications and risk factors in patients operated for a primary malignant brain tumor. Part II describes de development of a model for the prediction of individual-patient survival in glioblastoma patients. Part III encompasses the development of a natural language processing framework for automated medical text analysis. Machine learning algorithms should be considered as an extension to statistical approaches and exist along a continuum determined by how much is specified by humans and how much is learnt by the machine. Although machine learning algorithms can produce highly accurate predictions based on high-dimensional data, clinicians and researchers should interpret the clinical implications of these predictions on case-by-case basis. Show less
Pharmacogenomics (PGx) is widely recognized as an important aspect in personalizedMedicine. By analyzing and interpreting one’s genetic profile dose and drug adjustmentscan be made. In this way,... Show morePharmacogenomics (PGx) is widely recognized as an important aspect in personalizedMedicine. By analyzing and interpreting one’s genetic profile dose and drug adjustmentscan be made. In this way, one can strive to improve the safety and efficacy of drugtreatments. Nonetheless, not all genetic variability in drug response can be explained withcurrent PGx. In this thesis we explore the role of additional genetic factors which can explain this missing heritability. Firstly, rare and novel variants which are unaccounted for in routine PGx panels might play a role. Secondly, the complexity of pharmacogenes can result in an inability tounravel the genetic make-up of these genes. Thirdly, haplotype phasing is generally nottaken into account in PGx. Fourthly, all genetic variants are currently summarized intoone of four metabolic categories: poor metabolizers (PM), intermediate metabolizers(IM), normal metabolizers (NM) (previously EM) and ultra-rapid metabolizers (UM).However, enzyme activity is not a matter of ‘on’ or ‘off ’, but is more of a continuous scale.Finally, the effect of a genetic variant on drug metabolism shows substrate specific effects.This substrate specificity can result in erroneous extrapolation of variant effects to theentire range of substrates. The development of novel technologies to determine one’sgenetic make-up is evolving rapidly, thereby providing opportunities for the field of PGxto address these issues. In this thesis we show that by using long-read sequencing or trio-based sequencing more information can be obtained which can lead to a better understanding of the (rare) variants and can help with haplotype phasing. Moreover, we have shown that by combining long-read sequencing with artificial intelligence a substantial increase in explained variability can be achieved. Show less
Gitto, S.; Cuocolo, R.; Langevelde, K. van; Sande, M.A.J. van de; Parafioriti, A.; Luzzati, A.; ... ; Bloem, J.L. 2022
Background: Atypical cartilaginous tumour (ACT) and grade II chondrosarcoma (CS2) of long bones are respectively managed with watchful waiting or curettage and wide resection. Preoperatively,... Show moreBackground: Atypical cartilaginous tumour (ACT) and grade II chondrosarcoma (CS2) of long bones are respectively managed with watchful waiting or curettage and wide resection. Preoperatively, imaging diagnosis can be challenging due to interobserver variability and biopsy suffers from sample errors. The aim of this study is to determine diagnostic performance of MRI radiomics-based machine learning in differentiating ACT from CS2 of long bones. Methods: One-hundred-fifty-eight patients with surgically treated and histology-proven cartilaginous bone tumours were retrospectively included at two tertiary bone tumour centres. The training cohort consisted of 93 MRI scans from centre 1 (n=74 ACT; n=19 CS2). The external test cohort consisted of 65 MRI scans from centre 2 (n=45 ACT; n=20 CS2). Bidimensional segmentation was performed on T1-weighted MRI. Radiomic features were extracted. After dimensionality reduction and class balancing in centre 1, a machine-learning classifier (Extra Trees Classifier) was tuned on the training cohort using 10-fold cross-validation and tested on the external test cohort. In centre 2, its performance was compared with an experienced musculoskeletal oncology radiologist using McNemar's test. Findings: After tuning on the training cohort (AUC=0.88), the machine-learning classifier had 92% accuracy (60/ 65, AUC=0.94) in identifying the lesions in the external test cohort. Its accuracies in correctly classifying ACT and CS2 were 98% (44/45) and 80% (16/20), respectively. The radiologist had 98% accuracy (64/65) with no difference compared to the classifier (p=0.134). Interpretation: Machine learning showed high accuracy in classifying ACT and CS2 of long bones based on MRI radiomic features. Copyright (C) 2021 The Authors. Published by Elsevier B.V. Show less
Molecular imaging is one of the pillars of precision surgery. Its applications range from early diagnostics to therapy planning, execution, and the accurate assessment of outcomes. In particular,... Show moreMolecular imaging is one of the pillars of precision surgery. Its applications range from early diagnostics to therapy planning, execution, and the accurate assessment of outcomes. In particular, molecular imaging solutions are in high demand in minimally invasive surgical strategies, such as the substantially increasing field of robotic surgery. This review aims at connecting the molecular imaging and nuclear medicine community to the rapidly expanding armory of surgical medical devices. Such devices entail technologies ranging from artificial intelligence and computer-aided visualization technologies (software) to innovative molecular imaging modalities and surgical navigation (hardware). We discuss technologies based on their role at different steps of the surgical workflow, i.e., from surgical decision and planning, over to target localization and excision guidance, all the way to (back table) surgical verification. This provides a glimpse of how innovations from the technology fields can realize an exciting future for the molecular imaging and surgery communities. Show less
Maarseveen, T.D.; Maurits, M.P.; Niemantsverdriet, E.; Helm-van Mil, A.H.M. van der; Huizinga, T.W.J.; Knevel, R. 2021
Background Electronic health records (EHRs) offer a wealth of observational data. Machine-learning (ML) methods are efficient at data extraction, capable of processing the information-rich free... Show moreBackground Electronic health records (EHRs) offer a wealth of observational data. Machine-learning (ML) methods are efficient at data extraction, capable of processing the information-rich free-text physician notes in EHRs. The clinical diagnosis contained therein represents physician expert opinion and is more consistently recorded than classification criteria components. Objectives To investigate the overlap and differences between rheumatoid arthritis patients as identified either from EHR free-text through the extraction of the rheumatologist diagnosis using machine-learning (ML) or through manual chart-review applying the 1987 and 2010 RA classification criteria. Methods Since EHR initiation, 17,662 patients have visited the Leiden rheumatology outpatient clinic. For ML, we used a support vector machine (SVM) model to identify those who were diagnosed with RA by their rheumatologist. We trained and validated the model on a random selection of 2000 patients, balancing PPV and sensitivity to define a cutoff, and assessed performance on a separate 1000 patients. We then deployed the model on our entire patient selection (including the 3000). Of those, 1127 patients had both a 1987 and 2010 EULAR/ACR criteria status at 1 year after inclusion into the local prospective arthritis cohort. In these 1127 patients, we compared the patient characteristics of RA cases identified with ML and those fulfilling the classification criteria. Results The ML model performed very well in the independent test set (sensitivity=0.85, specificity=0.99, PPV=0.86, NPV=0.99). In our selection of patients with both EHR and classification information, 373 were recognized as RA by ML and 357 and 426 fulfilled the 1987 or 2010 criteria, respectively. Eighty percent of the ML-identified cases fulfilled at least one of the criteria sets. Both demographic and clinical parameters did not differ between the ML extracted cases and those identified with EULAR/ACR classification criteria. Conclusions With ML methods, we enable fast patient extraction from the huge EHR resource. Our ML algorithm accurately identifies patients diagnosed with RA by their rheumatologist. This resulting group of RA patients had a strong overlap with patients identified using the 1987 or 2010 classification criteria and the baseline (disease) characteristics were comparable. ML-assisted case labeling enables high-throughput creation of inclusive patient selections for research purposes. Show less
Diagnosis and treatment planning in ophthalmology heavily depend on clinical examination and advanced imaging modalities, which can be time-consuming and carry the risk of human error. Artificial... Show moreDiagnosis and treatment planning in ophthalmology heavily depend on clinical examination and advanced imaging modalities, which can be time-consuming and carry the risk of human error. Artificial intelligence (AI) and deep learning (DL) are being used in different fields of ophthalmology and in particular, when running diagnostics and predicting outcomes of anterior segment surgeries. This review will evaluate the recent developments in AI for diagnostics, surgical interventions, and prognosis of corneal diseases. It also provides a brief overview of the newer AI dependent modalities in corneal diseases. Show less
Zanca, F.; Hernandez-Giron, I.; Avanzo, M.; Guidi, G.; Crijns, W.; Diaz, O.; ... ; Kortesniemi, M. 2021
Purpose: To provide a guideline curriculum related to Artificial Intelligence (AI), for the education and training of European Medical Physicists (MPs).Materials and methods: The proposed... Show morePurpose: To provide a guideline curriculum related to Artificial Intelligence (AI), for the education and training of European Medical Physicists (MPs).Materials and methods: The proposed curriculum consists of two levels: Basic (introducing MPs to the pillars of knowledge, development and applications of AI, in the context of medical imaging and radiation therapy) and Advanced. Both are common to the subspecialties (diagnostic and interventional radiology, nuclear medicine, and radiation oncology). The learning outcomes of the training are presented as knowledge, skills and competences (KSC approach).Results: For the Basic section, KSCs were stratified in four subsections: (1) Medical imaging analysis and AI Basics; (2) Implementation of AI applications in clinical practice; (3) Big data and enterprise imaging, and (4) Quality, Regulatory and Ethical Issues of AI processes. For the Advanced section instead, a common block was proposed to be further elaborated by each subspecialty core curriculum. The learning outcomes were also translated into a syllabus of a more traditional format, including practical applications.Conclusions: This AI curriculum is the first attempt to create a guideline expanding the current educational framework for Medical Physicists in Europe. It should be considered as a document to top the sub-specialties' curriculums and adapted by national training and regulatory bodies. The proposed educational program can be implemented via the European School of Medical Physics Expert (ESMPE) course modules and - to some extent - also by the national competent EFOMP organizations, to reach widely the medical physicist community in Europe. Show less
Dotinga, M.; Dijk, J.D. van; Vendel, B.N.; Slump, C.H.; Portman, A.T.; Dalen, J.A. van 2021
Purpose Our aim was to develop and validate a machine learning (ML)-based approach for interpretation of I-123 FP-CIT SPECT scans to discriminate Parkinson's disease (PD) from non-PD and to... Show morePurpose Our aim was to develop and validate a machine learning (ML)-based approach for interpretation of I-123 FP-CIT SPECT scans to discriminate Parkinson's disease (PD) from non-PD and to determine its generalizability and clinical value in two centers.Methods We retrospectively included 210 consecutive patients who underwent I-123 FP-CIT SPECT imaging and had a clinically confirmed diagnosis. Linear support vector machine (SVM) was used to build a classification model to discriminate PD from non-PD based on I-123-FP-CIT striatal uptake ratios, age and gender of 90 patients. The model was validated on unseen data from the same center where the model was developed (n = 40) and consecutively on data from a different center (n = 80). Prediction performance was assessed and compared to the scan interpretation by expert physicians.Results Testing the derived SVM model on the unseen dataset (n = 40) from the same center resulted in an accuracy of 95.0%, sensitivity of 96.0% and specificity of 93.3%. This was identical to the classification accuracy of nuclear medicine physicians. The model was generalizable towards the other center as prediction performance did not differ thereby obtaining an accuracy of 82.5%, sensitivity of 88.5% and specificity of 71.4% (p = NS). This was comparable to that of nuclear medicine physicians (p = NS).Conclusion ML-based interpretation of I-123-FP-CIT scans results in accurate discrimination of PD from non-PD similar to visual assessment in both centers. The derived SVM model is therefore generalizable towards centers using comparable acquisition and image processing methods and implementation as diagnostic aid in clinical practice is encouraged. Show less
Purpose: To assess current perceptions, practices and education needs pertaining to artificial intelligence (AI) in the medical physics field.Methods: A web-based survey was distributed to the... Show morePurpose: To assess current perceptions, practices and education needs pertaining to artificial intelligence (AI) in the medical physics field.Methods: A web-based survey was distributed to the European Federation of Organizations for Medical Physics (EFOMP) through social media and email membership list. The survey included questions about education, personal knowledge, needs, research and professionalism around AI in medical physics. Demographics information were also collected. Responses were stratified and analysed by gender, type of institution and years of experience in medical physics. Statistical significance (p < 0.05) was assessed using paired t-test.Results: 219 people from 31 countries took part in the survey. 81% (n = 177) of participants agreed that AI will improve the daily work of Medical Physics Experts (MPEs) and 88% (n = 193) of respondents expressed the need for MPEs of specific training on AI. The average level of AI knowledge among participants was 2.3 +/- 1.0 (mean +/- standard deviation) in a 1-to-5 scale and 96% (n = 210) of participants showed interest in improving their AI skills. A significantly lower AI knowledge was observed for female participants (2.0 +/- 1.0), compared to male responders (2.4 +/- 1.0). 64% of participants indicated that they are not involved in AI projects. The percentage of female leading AI projects was significantly lower than the male counterparts (3% vs 19%).Conclusions: AI was perceived as a positive resource to support MPEs in their daily tasks. Participants demonstrated a strong interest in improving their current AI-related skills, enhancing the need for dedicated training for MPEs. Show less