European Radiology Experimental最新文献

Background: To retrospectively assess the added value of an artificial intelligence (AI) algorithm for detecting pulmonary nodules on ultra-low-dose computed tomography (ULDCT) performed at the emergency department (ED).

Methods: In the OPTIMACT trial, 870 patients with suspected nontraumatic pulmonary disease underwent ULDCT. The ED radiologist prospectively read the examinations and reported incidental pulmonary nodules requiring follow-up. All ULDCTs were processed post hoc using an AI deep learning software marking pulmonary nodules ≥ 6 mm. Three chest radiologists independently reviewed the subset of ULDCTs with either prospectively detected incidental nodules in 35/870 patients or AI marks in 458/870 patients; findings scored as nodules by at least two chest radiologists were used as true positive reference standard. Proportions of true and false positives were compared.

Results: During the OPTIMACT study, 59 incidental pulmonary nodules requiring follow-up were prospectively reported. In the current analysis, 18/59 (30.5%) nodules were scored as true positive while 104/1,862 (5.6%) AI marks in 84/870 patients (9.7%) were scored as true positive. Overall, 5.8 times more (104 versus 18) true positive pulmonary nodules were detected with the use of AI, at the expense of 42.9 times more (1,758 versus 41) false positives. There was a median number of 1 (IQR: 0-2) AI mark per ULDCT.

Conclusion: The use of AI on ULDCT in patients suspected of pulmonary disease in an emergency setting results in the detection of many more incidental pulmonary nodules requiring follow-up (5.8×) with a high trade-off in terms of false positives (42.9×).

Relevance statement: AI aids in the detection of incidental pulmonary nodules that require follow-up at chest-CT, aiding early pulmonary cancer detection but also results in an increase of false positive results that are mainly clustered in patients with major abnormalities.

Trial registration: The OPTIMACT trial was registered on 6 December 2016 in the National Trial Register (number NTR6163) (onderzoekmetmensen.nl).

Key points: An AI deep learning algorithm was tested on 870 ULDCT examinations acquired in the ED. AI detected 5.8 times more pulmonary nodules requiring follow-up (true positives). AI resulted in the detection of 42.9 times more false positive results, clustered in patients with major abnormalities. AI in the ED setting may aid in early pulmonary cancer detection with a high trade-off in terms of false positives.

Background: To investigate changes in pulmonary artery pressure (PAP), blood indices, and myocardial microcirculation in rats returning from high altitude (HA) to moderate altitude (MA).

Methods: Forty 4-week-old male Sprague-Dawley rats were randomly divided into four groups with ten rats in each group. One group was transported to the MA area (MA-group), and the other three groups were transported to HA (HA-group-A, HA-group-B, and HA-group-C). After 28 weeks of age, the rats from the HA area were transported to the MA area for 0 days, 10 days, and 20 days, respectively. PAP, routine blood tests, and computed tomography myocardial perfusion indices were measured.

Results: Compared with the MA-group, the body weight of HA-groups decreased (p < 0.05), and PAP in HA-group-A and HA-group-B increased (p < 0.05). In the HA groups, PAP initially increased and then decreased. Compared with the MA-group, red blood cells (RBC), hemoglobin (HGB), and hematocrit (HCT) of rats in HA-group-A increased (p < 0.05). Compared with the HA-group-A, RBC, HGB, and HCT of HA-group-B gradually decreased (p < 0.05) while MCV decreased (p < 0.05), and PLT of HA-group-C increased (p < 0.05). Compared with the MA group, blood flow (BF) and blood volume (BV) of the HA-group-A decreased (p < 0.05). Compared with the HA-group-A, TTP increased first and then decreased (p < 0.05), and BF and BV increased gradually (p < 0.05). Pathological results showed that myocardial fiber arrangement was disordered, and cell space widened in the HA group.

Conclusion: PAP, blood parameters, and myocardial microcirculation in rats returning from high to MA exhibited significant changes.

Relevance statement: This study provides an experimental basis for understanding the physiological and pathological mechanisms during the process of deacclimatization to HA and offers new insights for the prevention and treatment of deacclimatization to HA syndrome.

Key points: Forty rats were raised in a real plateau environment. Myocardial microcirculation was detected by CT myocardial perfusion imaging. The PAP of the unacclimated rats increased first and then decreased. The myocardial microcirculation of the deacclimated rats showed hyperperfusion changes.

Feature extraction and selection from medical data are the basis of radiomics and image biomarker discovery for various architectures, including convolutional neural networks (CNNs). We herein describe the typical radiomics steps and the components of a CNN for both deep feature extraction and end-to-end approaches. We discuss the curse of dimensionality, along with dimensionality reduction techniques. Despite the outstanding performance of deep learning (DL) approaches, the use of handcrafted features instead of deep learned features needs to be considered for each specific study. Dataset size is a key factor: large-scale datasets with low sample diversity could lead to overfitting; limited sample sizes can provide unstable models. The dataset must be representative of all the "facets" of the clinical phenomenon/disease investigated. The access to high-performance computational resources from graphics processing units is another key factor, especially for the training phase of deep architectures. The advantages of multi-institutional federated/collaborative learning are described. When large language models are used, high stability is needed to avoid catastrophic forgetting in complex domain-specific tasks. We highlight that non-DL approaches provide model explainability superior to that provided by DL approaches. To implement explainability, the need for explainable AI arises, also through post hoc mechanisms. RELEVANCE STATEMENT: This work aims to provide the key concepts for processing the imaging features to extract reliable and robust image biomarkers. KEY POINTS: The key concepts for processing the imaging features to extract reliable and robust image biomarkers are provided. The main differences between radiomics and representation learning approaches are highlighted. The advantages and disadvantages of handcrafted versus learned features are given without losing sight of the clinical purpose of artificial intelligence models.

Background: The presence of a blurred area, depending on its localization, in a mammogram can limit diagnostic accuracy. The goal of this study was to develop a model for automatic detection of blur in diagnostically relevant locations in digital mammography.

Methods: A retrospective dataset consisting of 152 examinations acquired with mammography machines from three different vendors was utilized. The blurred areas were contoured by expert breast radiologists. Normalized Wiener spectra (nWS) were extracted in a sliding window manner from each mammogram. These spectra served as input for a convolutional neural network (CNN) generating the probability of the spectra originating from a blurred region. The resulting blur probability mask, upon thresholding, facilitated the classification of a mammogram as either blurred or sharp. Ground truth for the test set was defined by the consensus of two radiologists.

Results: A significant correlation between the view (p < 0.001), as well as between the laterality and the presence of blur (p = 0.004) was identified. The developed model AUROC of 0.808 (95% confidence interval 0.794-0.821) aligned with the consensus in 78% (67-83%) of mammograms classified as blurred. For mammograms classified by consensus as sharp, the model achieved agreement in 75% (67-83%) of them.

Conclusion: A model for blur detection was developed and assessed. The results indicate that a robust approach to blur detection, based on feature extraction in frequency space, tailored to radiologist expertise regarding clinical relevance, could eliminate the subjectivity associated with the visual assessment.

Relevance statement: This blur detection model, if implemented in clinical practice, could provide instantaneous feedback to technicians, allowing for prompt mammogram retakes and ensuring that only high-quality mammograms are sent for screening and diagnostic tasks.

Key points: Blurring in mammography limits radiologist interpretation and diagnostic accuracy. This objective blur detection tool ensures image quality, and reduces retakes and unnecessary exposures. Wiener spectrum analysis and CNN enabled automated blur detection in mammography.

Background: To investigate the artifact sizes of four common breast clip-markers on a standard breast magnetic resonance imaging (MRI) protocol in an in vitro phantom model.

Methods: Using 1.5-T and 3-T whole-body scanners with an 18-channel breast coil, artifact dimensions of four breast biopsy markers in an agarose-gel phantom were measured by two readers on images obtained with the following sequences: T2-weighted fast spin-echo short inversion time fat-suppressed inversion-recovery with magnitude reconstruction (T2-TIRM); T1-weighted spoiled gradient-echo with fat suppression (T1_FL3D), routinely used for dynamic contrast-enhanced imaging; diffusion-weighted imaging (DWI), including a readout segmented echo-planar imaging (RESOLVE-DWI) and echo-planar imaging sequence (EPI-DWI). After outlining the artifacts by freehand regions of interest, sagittal and lateral diameters in axial images were measured.

Results: Interreader agreement for artifact size quantification was high, depending on the sequence (80.4-94.8%). Overall, the size, shape, and appearance of artifacts depended on clip type and MRI sequence. The artifact size ranged from 5.7 × 8.5 mm² to 13.4 × 17.7 mm² at 1.5 T and from 6.6 × 8.2 mm² to 17.7 × 20.7 mm² at 3 T. Clip artifacts were largest on EPI-DWI and RESOLVE-DWI (p ≤ 0.016). In three out of four clips, T2-TIRM showed the smallest artifact (p ≤ 0.002), while in one clip the artifact was smallest on T1_FL3D (p = 0.026). With the exception of one clip in the RESOLVE sequence, all clips showed a decrease in the artifact area from DWI to ADC images (p ≤ 0.037).

Conclusion: Breast clip-marker MRI artifact appearances depend on clip type, field strength, and sequence and may reach a significant size, potentially obscuring smaller lesions and hindering accurate assessment of breast tumors.

Relevance statement: Considerable variations in artifact size and characteristics across different breast clips, MRI sequences, and field strengths exist. Awareness of these artifacts and their characteristics is essential to ensure accurate interpretation of scans and appropriate treatment planning.

Key points: Awareness of breast clip artifacts is essential for accurate interpretation of MRI. The appearance of artifacts depends on breast clip type, field strength, and sequence. Clip-related artifacts might hinder the visibility of small lesions.

Background: Dual-energy computed tomography (DECT) is a noninvasive diagnostic tool for gouty arthritis. This study aimed to compare two postprocessing techniques for monosodium urate (MSU) detection: conventional two-material decomposition and material map-based decomposition.

Methods: A raster phantom and an ex vivo biophantom, embedded with four different MSU concentrations, were scanned in two high-end CT scanners. Scanner 1 used the conventional postprocessing method while scanner 2 employed the material map approach. Volumetric analysis was performed to determine MSU detection, and image quality parameters, such as signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR), were computed.

Results: The material map-based method demonstrated superior MSU detection. Specifically, scanner 2 yielded total MSU volumes of 5.29 ± 0.28 mL and 4.52 ± 0.29 mL (mean ± standard deviation) in the raster and biophantom, respectively, versus 2.35 ± 0.23 mL and 1.15 ± 0.17 mL for scanner 1. Radiation dose correlated positively with detection for the conventional scanner, while there was no such correlation for the material map-based decomposition method in the biophantom. Despite its higher detection rate, material map-based decomposition was inferior in terms of SNR, CNR, and artifacts.

Conclusion: While material map-based decomposition resulted in superior MSU detection, it is limited by challenges such as increased artifacts. Our findings highlight the potential of this method for gout diagnosis while underscoring the need for further research to enhance its clinical reliability.

Relevance statement: Advanced postprocessing such as material-map-based two-material decomposition might improve the sensitivity for gouty arthritis in clinical practice, thus, allowing for lower radiation doses or better sensitivity for gouty tophi.

Key points: Dual-energy CT showed limited sensitivity for tophi with low MSU concentrations. Materiel-map-based decomposition increased sensitivity compared to conventional two-material decomposition. The advantages of material-map-based decomposition outweigh lower image quality and increased artifact load.

Background: We developed a framework for segmenting and modeling lenticulostriate arteries (LSAs) on 7-T time-of-flight magnetic resonance angiography and tested its performance on cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL) patients and controls.

Methods: We prospectively included 29 CADASIL patients and 21 controls. The framework includes a small-patch convolutional neural network (SP-CNN) for fine segmentation, a random forest for modeling LSAs, and a screening model for removing wrong branches. The segmentation performance of our SP-CNN was compared to competitive networks. External validation with different resolution was performed on ten patients with aneurysms. Dice similarity coefficient (DSC) and Hausdorff distance (HD) between each network and manual segmentation were calculated. The modeling results of the centerlines, diameters, and lengths of LSAs were compared against manual labeling by four neurologists.

Results: The SP-CNN achieved higher DSC (92.741 ± 2.789, mean ± standard deviation) and lower HD (0.610 ± 0.141 mm) in the segmentation of LSAs. It also outperformed competitive networks in the external validation (DSC 82.6 ± 5.5, HD 0.829 ± 0.143 mm). The framework versus manual difference was lower than the manual inter-observer difference for the vessel length of primary branches (median -0.040 mm, interquartile range -0.209 to 0.059 mm) and secondary branches (0.202 mm, 0.016-0.537 mm), as well as for the offset of centerlines of primary branches (0.071 mm, 0.065-0.078 mm) and secondary branches (0.072, 0.064-0.080 mm), with p < 0.001 for all comparisons.

Conclusion: Our framework for LSAs modeling/quantification demonstrated high reliability and accuracy when compared to manual labeling.

Trial registration: NCT05902039 ( https://clinicaltrials.gov/study/NCT05902039?cond=NCT05902039 ).

Relevance statement: The proposed automatic segmentation and modeling framework offers precise quantification of the morphological parameters of lenticulostriate arteries. This innovative technology streamlines diagnosis and research of cerebral small vessel disease, eliminating the burden of manual labeling, facilitating cohort studies and clinical diagnosis.

Key points: The morphology of LSAs is important in the diagnosis of CSVD but difficult to quantify. The proposed algorithm achieved the performance equivalent to manual labeling by neurologists. Our method can provide standardized quantitative results, reducing radiologists' workload in cohort studies.

Background: Dark-field radiography imaging exploits the wave character of x-rays to measure small-angle scattering on material interfaces, providing structural information with low radiation exposure. We explored the potential of dark-field imaging of bone microstructure to improve the assessment of bone strength in osteoporosis.

Methods: We prospectively examined 14 osteoporotic/osteopenic and 21 non-osteoporotic/osteopenic human cadaveric vertebrae (L2-L4) with a clinical dark-field radiography system, micro-computed tomography (CT), and spectral CT. Dark-field images were obtained in both vertical and horizontal sample positions. Bone microstructural parameters (trabecular number, Tb.N; trabecular thickness, Tb.Th; bone volume fraction, BV/TV; degree of anisotropy, DA) were measured using standard ex vivo micro-CT, while hydroxyapatite density was measured using spectral CT. Correlations were assessed using Spearman rank correlation coefficients.

Results: The measured dark-field signal was lower in osteoporotic/osteopenic vertebrae (vertical position, 0.23 ± 0.05 versus 0.29 ± 0.04, p < 0.001; horizontal position, 0.28 ± 0.06 versus 0.34 ± 0.04, p = 0.003). The dark-field signal from the vertical position correlated significantly with Tb.N (ρ = 0.46, p = 0.005), BV/TV (ρ = 0.45, p = 0.007), DA (ρ = -0.43, p = 0.010), and hydroxyapatite density (ρ = 0.53, p = 0.010). The calculated ratio of vertical/horizontal dark-field signal correlated significantly with Tb.N (ρ = 0.43, p = 0.011), BV/TV (ρ = 0.36, p = 0.032), DA (ρ = -0.51, p = 0.002), and hydroxyapatite density (ρ = 0.42, p = 0.049).

Conclusion: Dark-field radiography is a feasible modality for drawing conclusions on bone microarchitecture in human cadaveric vertebral bone.

Relevance statement: Gaining knowledge of the microarchitecture of bone contributes crucially to predicting bone strength in osteoporosis. This novel radiographic approach based on dark-field x-rays provides insights into bone microstructure at a lower radiation exposure than that of CT modalities.

Key points: Dark-field radiography can give information on bone microstructure with low radiation exposure. The dark-field signal correlated positively with bone microstructure parameters. Dark-field signal correlated negatively with the degree of anisotropy. Dark-field radiography helps to determine the directionality of trabecular loss.

Background: To assess the reliability and comprehensibility of breast radiology reports simplified by artificial intelligence using the large language model (LLM) ChatGPT-4o.

Methods: A radiologist with 20 years' experience selected 21 anonymized breast radiology reports, 7 mammography, 7 breast ultrasound, and 7 breast magnetic resonance imaging (MRI), categorized according to breast imaging reporting and data system (BI-RADS). These reports underwent simplification by prompting ChatGPT-4o with "Explain this medical report to a patient using simple language". Five breast radiologists assessed the quality of these simplified reports for factual accuracy, completeness, and potential harm with a 5-point Likert scale from 1 (strongly agree) to 5 (strongly disagree). Another breast radiologist evaluated the text comprehension of five non-healthcare personnel readers using a 5-point Likert scale from 1 (excellent) to 5 (poor). Descriptive statistics, Cronbach's α, and the Kruskal-Wallis test were used.

Results: Mammography, ultrasound, and MRI showed high factual accuracy (median 2) and completeness (median 2) across radiologists, with low potential harm scores (median 5); no significant group differences (p ≥ 0.780), and high internal consistency (α > 0.80) were observed. Non-healthcare readers showed high comprehension (median 2 for mammography and MRI and 1 for ultrasound); no significant group differences across modalities (p = 0.368), and high internal consistency (α > 0.85) were observed. BI-RADS 0, 1, and 2 reports were accurately explained, while BI-RADS 3-6 reports were challenging.

Conclusion: The model demonstrated reliability and clarity, offering promise for patients with diverse backgrounds. LLMs like ChatGPT-4o could simplify breast radiology reports, aid in communication, and enhance patient care.

Relevance statement: Simplified breast radiology reports generated by ChatGPT-4o show potential in enhancing communication with patients, improving comprehension across varying educational backgrounds, and contributing to patient-centered care in radiology practice.

Key points: AI simplifies complex breast imaging reports, enhancing patient understanding. Simplified reports from AI maintain accuracy, improving patient comprehension significantly. Implementing AI reports enhances patient engagement and communication in breast imaging.

Background: Our aim was to train and test a deep learning-based algorithm for automatically segmenting kidneys and renal cysts in patients with autosomal dominant polycystic kidney disease (ADPKD).

Methods: We retrospectively selected all ADPKD patients who underwent renal MRI with coronal T2-weighted imaging at our institution from 2008 to 2022. The 20 most recent examinations constituted the test dataset, to mimic pseudoprospective enrolment. The remaining ones constituted the training dataset to which eight normal renal MRIs were added. Kidneys and cysts ground truth segmentations were performed on coronal T2-weighted images by a junior radiologist supervised by an experienced radiologist. Kidneys and cysts of the 20 test MRIs were segmented by the algorithm and three independent human raters. Segmentations were compared using overlap metrics. The total kidney volume (TKV), total cystic volume (TCV), and cystic index (TCV divided by TKV) were compared using Bland-Altman analysis.

Results: We included 164 ADPKD patients. Dice similarity coefficients ranged from 85.9% to 87.4% between the algorithms and the raters' segmentations and from 84.2% to 86.2% across raters' segmentations. For TCV assessment, the biases ± standard deviations (SD) were 3-19 ± 137-151 mL between the algorithm and the raters, and 22-45 ± 49-57 mL across raters. The algorithm underestimated TKV and TCV in two outliers with TCV > 2800 mL. For cystic index assessment, the biases ± SD were 2.5-6.9% ± 6.7-8.3% between the algorithm and the raters, and 2.1-9.4 ± 7.4-11.6% across raters.

Conclusion: The algorithm's performance fell within the range of inter-rater variability, but large TKV and TCV were underestimated.

Relevance statement: Accurate automated segmentation of the renal cysts will enable the large-scale evaluation of the prognostic value of TCV and cystic index in ADPKD patients. If these biomarkers are prognostic, then automated segmentation will facilitate their use in daily routine.

Key points: Cystic volume is an emerging biomarker in ADPKD. The algorithm's performance in segmenting kidneys and cysts fell within interrater variability. The segmentation of very large cysts, under-represented in the training dataset, needs improvement.