Explore the vast range of titles available.

Artificial intelligence (AI) systems for computer-aided diagnosis and image-based screening are being adopted worldwide by medical institutions. In such a context, generating fair and unbiased classifiers becomes of paramount importance. The research community of medical image computing is making great efforts in developing more accurate algorithms to assist medical doctors in the difficult task of disease diagnosis. However, little attention is paid to the way databases are collected and how this may influence the performance of AI systems. Our study sheds light on the importance of gender balance in medical imaging datasets used to train AI systems for computer-assisted diagnosis. We provide empirical evidence supported by a large-scale study, based on three deep neural network architectures and two well-known publicly available X-ray image datasets used to diagnose various thoracic diseases under different gender imbalance conditions. We found a consistent decrease in performance for underrepresented genders when a minimum balance is not fulfilled. This raises the alarm for national agencies in charge of regulating and approving computer-assisted diagnosis systems, which should include explicit gender balance and diversity recommendations. We also establish an open problem for the academic medical image computing community which needs to be addressed by novel algorithms endowed with robustness to gender imbalance.

PurposeTo assess the dose performance in terms of image quality of filtered back projection (FBP) and two generations of iterative reconstruction (IR) algorithms developed by the most common CT vendors.Materials and methodsWe used four CT systems equipped with a hybrid/statistical IR (H/SIR) and a full/partial/advanced model-based IR (MBIR) algorithms. Acquisitions were performed on an ACR phantom at five dose levels. Raw data were reconstructed using a standard soft tissue kernel for FBP and one iterative level of the two IR algorithm generations. The noise power spectrum (NPS) and the task-based transfer function (TTF) were computed. A detectability index (d′) was computed to model the detection task of a large mass in the liver (large feature; 120 HU and 25-mm diameter) and a small calcification (small feature; 500 HU and 1.5-mm diameter).ResultsWith H/SIR, the highest values of d′ for both features were found for Siemens, then for Canon and the lowest values for Philips and GE. For the large feature, potential dose reductions with MBIR compared with H/SIR were − 35% for GE, − 62% for Philips, and − 13% for Siemens; for the small feature, corresponding reductions were − 45%, − 78%, and − 14%, respectively. With the Canon system, a potential dose reduction of − 32% was observed only for the small feature with MBIR compared with the H/SIR algorithm. For the large feature, the dose increased by 100%.ConclusionThis multivendor comparison of several versions of IR algorithms allowed to compare the different evolution within each vendor. The use of d′ is highly adapted and robust for an optimization process.Key Points• The performance of four CT systems was evaluated by using imQuest software to assess noise characteristic, spatial resolution, and lesion detection.• Two task functions were defined to model the detection task of a large mass in the liver and a small calcification.• The advantage of task-based image quality assessment for radiologists is that it does not include only complicated metrics, but also clinically meaningful image quality.

There have been few independent evaluations of computer-aided detection (CAD) software for tuberculosis (TB) screening, despite the rapidly expanding array of available CAD solutions. We developed a test library of chest X-ray (CXR) images which was blindly re-read by two TB clinicians with different levels of experience and then processed by 12 CAD software solutions. Using Xpert MTB/RIF results as the reference standard, we compared the performance characteristics of each CAD software against both an Expert and Intermediate Reader, using cut-off thresholds which were selected to match the sensitivity of each human reader. Six CAD systems performed on par with the Expert Reader (Qure.ai, DeepTek, Delft Imaging, JF Healthcare, OXIPIT, and Lunit) and one additional software (Infervision) performed on par with the Intermediate Reader only. Qure.ai, Delft Imaging and Lunit were the only software to perform significantly better than the Intermediate Reader. The majority of these CAD software showed significantly lower performance among participants with a past history of TB. The radiography equipment used to capture the CXR image was also shown to affect performance for some CAD software. TB program implementers now have a wide selection of quality CAD software solutions to utilize in their CXR screening initiatives.

Chest radiography is the most common diagnostic imaging examination performed in emergency departments (EDs). Augmenting clinicians with automated preliminary read assistants could help expedite their workflows, improve accuracy, and reduce the cost of care. To assess the performance of artificial intelligence (AI) algorithms in realistic radiology workflows by performing an objective comparative evaluation of the preliminary reads of anteroposterior (AP) frontal chest radiographs performed by an AI algorithm and radiology residents. This diagnostic study included a set of 72 findings assembled by clinical experts to constitute a full-fledged preliminary read of AP frontal chest radiographs. A novel deep learning architecture was designed for an AI algorithm to estimate the findings per image. The AI algorithm was trained using a multihospital training data set of 342 126 frontal chest radiographs captured in ED and urgent care settings. The training data were labeled from their associated reports. Image-based F1 score was chosen to optimize the operating point on the receiver operating characteristics (ROC) curve so as to minimize the number of missed findings and overcalls per image read. The performance of the model was compared with that of 5 radiology residents recruited from multiple institutions in the US in an objective study in which a separate data set of 1998 AP frontal chest radiographs was drawn from a hospital source representative of realistic preliminary reads in inpatient and ED settings. A triple consensus with adjudication process was used to derive the ground truth labels for the study data set. The performance of AI algorithm and radiology residents was assessed by comparing their reads with ground truth findings. All studies were conducted through a web-based clinical study application system. The triple consensus data set was collected between February and October 2018. The comparison study was preformed between January and October 2019. Data were analyzed from October to February 2020. After the first round of reviews, further analysis of the data was performed from March to July 2020. The learning performance of the AI algorithm was judged using the conventional ROC curve and the area under the curve (AUC) during training and field testing on the study data set. For the AI algorithm and radiology residents, the individual finding label performance was measured using the conventional measures of label-based sensitivity, specificity, and positive predictive value (PPV). In addition, the agreement with the ground truth on the assignment of findings to images was measured using the pooled κ statistic. The preliminary read performance was recorded for AI algorithm and radiology residents using new measures of mean image-based sensitivity, specificity, and PPV designed for recording the fraction of misses and overcalls on a per image basis. The 1-sided analysis of variance test was used to compare the means of each group (AI algorithm vs radiology residents) using the F distribution, and the null hypothesis was that the groups would have similar means. The trained AI algorithm achieved a mean AUC across labels of 0.807 (weighted mean AUC, 0.841) after training. On the study data set, which had a different prevalence distribution, the mean AUC achieved was 0.772 (weighted mean AUC, 0.865). The interrater agreement with ground truth finding labels for AI algorithm predictions had pooled κ value of 0.544, and the pooled κ for radiology residents was 0.585. For the preliminary read performance, the analysis of variance test was used to compare the distributions of AI algorithm and radiology residents' mean image-based sensitivity, PPV, and specificity. The mean image-based sensitivity for AI algorithm was 0.716 (95% CI, 0.704-0.729) and for radiology residents was 0.720 (95% CI, 0.709-0.732) (P = .66), while the PPV was 0.730 (95% CI, 0.718-0.742) for the AI algorithm and 0.682 (95% CI, 0.670-0.694) for the radiology residents (P < .001), and specificity was 0.980 (95% CI, 0.980-0.981) for the AI algorithm and 0.973 (95% CI, 0.971-0.974) for the radiology residents (P < .001). These findings suggest that it is possible to build AI algorithms that reach and exceed the mean level of performance of third-year radiology residents for full-fledged preliminary read of AP frontal chest radiographs. This diagnostic study also found that while the more complex findings would still benefit from expert overreads, the performance of AI algorithms was associated with the amount of data available for training rather than the level of difficulty of interpretation of the finding. Integrating such AI systems in radiology workflows for preliminary interpretations has the potential to expedite existing radiology workflows and address resource scarcity while improving overall accuracy and reducing the cost of care.

Objective To investigate the feasibility of standardizing RT simulation CT scanner protocols between vendors using target‐based image quality (IQ) metrics. Method and materials A systematic assessment process in phantom was developed to standardize clinical scan protocols for scanners from different vendors following these steps: (a) images were acquired by varying CTDIvol and using an iterative reconstruction (IR) method (IR: iDose and model‐based iterative reconstruction [IMR] of CTp‐Philips Big Bore scanner, SAFIRE of CTs‐Siemens biograph PETCT scanner), (b) CT exams were classified into body and brain protocols, (c) the rescaled noise power spectrum (NPS) was calculated, (d) quantified the IQ change due to varied CTDIvol and IR, and (e) matched the IR strength level. IQ metrics included noise and texture from NPS, contrast, and contrast‐to‐noise ratio (CNR), low contrast detectability (d′). Area under curve (AUC) of the receiver operation characteristic curve of d′ was calculated and compared. Results The level of change in the IQ ratio was significant (>0.6) when using IMR. The IQ ratio change was relatively low to moderate when using either iDose in CTp (0.1–0.5) or SAFIRE in CTs (0.1–0.6). SAFIRE‐2 in CTs showed a closer match to the reference body protocol when compared to iDose‐3 in CTp. In the brain protocol, iDose‐3 in CTp could be matched to the low to moderate level of SAFIRE in CTs. The AUC of d′ was highest when using IMR in CTp with lower CTDIvol, and SAFIRE in CTs performed better than iDose in CTp Conclusion It is possible to use target‐based IQ metrics to evaluate the performance of the system and operations across various scanners in a phantom. This can serve as an initial reference to convert clinical scanned protocols from one CT simulation scanner to another.

Since the invention of Computed Tomography (CT), many technological advances emerged to improve the image sensitivity and resolution. However, no new source types were developed for clinical use. In this study, for the first time, coherent monochromatic X-rays from a synchrotron radiation source were used to acquire 3D CTs on patients. The aim of this work was to evaluate the clinical potential of the images acquired using Synchrotron Radiation CT (SRCT). SRCTs were acquired using monochromatic X-rays tuned at 80 keV (0.350 × 0.350 × 2 mm 3 voxel size). A quantitative image quality comparison study was carried out on phantoms between a state of the art clinical CT and SRCT images. Dedicated iterative algorithms were developed to optimize the image quality and further reduce the delivered dose by a factor of 12 while keeping a better image quality than the one obtained with a clinical CT scanner. We finally show in this paper the very first SRCT results of one patient who received Synchrotron Radiotherapy in an ongoing clinical trial. This demonstrates the potential of the technique in terms of image quality improvement at a reduced radiation dose for inner ear visualization.

To evaluate artifact reduction by virtual monoenergetic images (VMI) and metal artifact reduction algorithms (MAR) as well as the combination of both approaches (VMI MAR ) compared to conventional CT images (CI) as standard of reference. In this retrospective study, 35 patients were included who underwent spectral-detector CT (SDCT) with additional MAR-reconstructions due to artifacts from coils or clips. CI, VMI, MAR and VMI MAR (range: 100–200 keV, 10 keV-increment) were reconstructed. Region-of-interest based objective analysis was performed by assessing mean and standard deviation of attenuation (HU) in hypo- and hyperdense artifacts from coils and clips. Visually, extent of artifact reduction and diagnostic assessment were rated. Compared to CI, VMI ≥ 100 keV, MAR and VMI MAR between 100–200 keV increased attenuation in hypoattenuating artifacts (CI/VMI 200keV /MAR/VMI MAR200keV , HU: −77.6 ± 81.1/−65.1 ± 103.2/−36.9 ± 27.7/−21.1 ± 26.7) and decreased attenuation in hyperattenuating artifacts (HU: 47.4 ± 32.3/42.1 ± 50.2/29.5 ± 18.9/20.8 ± 25.8). However, differences were only significant for MAR in hypodense and VMI MAR in hypo- and hyperdense artifacts (p < 0.05). Visually, hypo- and hyperdense artifacts were significantly reduced compared to CI by VMI ≥140/100keV , MAR and VMI MAR≥100keV . Diagnostic assessment of surrounding brain tissue was significantly improved in VMI ≥100keV , MAR and VMI MAR≥100keV . The combination of VMI and MAR facilitates a significant reduction of artifacts adjacent to intracranial coils and clips. Hence, if available, these techniques should be combined for optimal reduction of artifacts following intracranial aneurysm treatment.

ObjectivesTo evaluate the image quality of ultralow-dose computed tomography (ULDCT) reconstructed with knowledge-based iterative model reconstruction (IMR) in patients with pulmonary tuberculosis (TB).MethodsThis IRB-approved prospective study enrolled 59 consecutive patients (mean age, 43.9 ± 16.6 years; F:M 18:41) with known or suspected pulmonary TB. Patients underwent a low-dose CT (LDCT) using automatic tube current modulation followed by an ULDCT using fixed tube current. Raw image data were reconstructed with filtered-back projection (FBP), hybrid iterative reconstruction (iDose), and IMR. Objective measurements including CT attenuation, image noise, and contrast-to-noise ratio (CNR) were assessed and compared using repeated-measures analysis of variance. Overall image quality and visualization of normal and pathological findings were subjectively scored on a five-point scale. Radiation output and subjective scores were compared by the paired Student t test and Wilcoxon signed-rank test, respectively.ResultsCompared with FBP and iDose, IMR yielded significantly lower noise and higher CNR values at both dose levels (p < 0.01). Subjective ratings for pathological findings including centrilobular nodules, consolidation, tree-in-bud, and cavity were significantly better with ULDCT IMR images than those with LDCT iDose images (p < 0.01), but blurred edges were observed. With IMR implementation, a 59% reduction of the mean effective dose was achieved with ULDCT (0.28 ± 0.02 mSv) compared with LDCT (0.69 ± 0.15 mSv) without impairing image quality (p < 0.001).ConclusionsIMR offers considerable noise reduction and improvement in image quality for patients with pulmonary TB undergoing chest ULDCT at an effective dose of 0.28 mSv.Key Points• Radiation dose is a major concern for tuberculosis patients requiring repeated follow-up CT.• IMR allows substantial radiation dose reduction in chest CT without compromising image quality.• ULDCT reconstructed with IMR allows accurate depiction of CT features of pulmonary tuberculosis.

Myocardial deformation is a sensitive marker of sub-clinical myocardial dysfunction that carries independent prognostic significance across a broad range of cardiovascular diseases. It is now possible to perform 3D feature tracking of SSFP cines on cardiac magnetic resonance imaging (FT-CMR). This study provides reference ranges for 3D FT-CMR and assesses its reproducibility compared to 2D FT-CMR. One hundred healthy individuals with 10 men and women in each of 5 age deciles from 20 to 70 years, underwent 2D and 3D FT-CMR of left ventricular myocardial strain and strain rate using SSFP cines. Good health was defined by the absence of hypertension, diabetes, obesity, dyslipidaemia, or any cardiovascular, renal, hepatic, haematological and systemic inflammatory disease. Normal values for myocardial strain assessed by 3D FT-CMR were consistently lower compared with 2D FT-CMR measures [global circumferential strain (GCS) 3D − 17.6 ± 2.6% vs. 2D − 20.9 ± 3.7%, P < 0.005]. Validity of 3D FT-CMR was confirmed against other markers of systolic function. The 3D algorithm improved reproducibility compared to 2D, with GCS having the best inter-observer agreement [intra-class correlation (ICC) 0.88], followed by global radial strain (GRS; ICC 0.79) and global longitudinal strain (GLS, ICC 0.74). On linear regression analyses, increasing age was weakly associated with increased GCS (R2 = 0.15, R = 0.38), peak systolic strain rate, peak late diastolic strain rate, and lower peak early systolic strain rate. 3D FT-CMR offers superior reproducibility compared to 2D FT-CMR, with circumferential strain and strain rates offering excellent intra- and inter-observer variability. Normal range values for myocardial strain measurements using 3D FT-CMR are provided.

Purpose To evaluate the subjective and objective qualities of computed tomography (CT) venography images at 80 kVp using model-based iterative reconstruction (MBIR) and to compare these with those of filtered back projection (FBP) and adaptive statistical iterative reconstruction (ASIR) using the same CT data sets. Materials and methods Forty-four patients (mean age: 56.1 ± 18.1) who underwent 80 kVp CT venography (CTV) for the evaluation of deep vein thrombosis (DVT) during 4 months were enrolled in this retrospective study. The same raw data were reconstructed using FBP, ASIR, and MBIR. Objective and subjective image analysis were performed at the inferior vena cava (IVC), femoral vein, and popliteal vein. Results The mean CNR of MBIR was significantly greater than those of FBP and ASIR and images reconstructed using MBIR had significantly lower objective image noise (p < .001). Subjective image quality and confidence of detecting DVT by MBIR group were significantly greater than those of FBP and ASIR (p < .005), and MBIR had the lowest score for subjective image noise (p < .001). Conclusion CTV at 80 kVp with MBIR was superior to FBP and ASIR regarding subjective and objective image qualities. Key Points • MBIR provides superior image quality compared with FBP and ASIR • CTV at 80kVp with MBIR improves diagnostic confidence in diagnosing DVT • CTV at 80kVp with MBIR presents better image quality with low radiation