Iwona Kucybała
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Iwona Kucybała has a total of 57 co-authors. Here are the top co-authors by number of shared publications:
- Andrzej Urbanik
Department of Diagnostic Imaging, Jagiellonian University Medical College, 30-663 Krakow, Poland
7 shared publications - Wadim Wojciechowski
Department of Radiology, Jagiellonian University Medical College, ul. Botaniczna 3, 31-503 Krakow, Poland
5 shared publications - Zbisław Tabor
Faculty of Electrical Engineering, Automatics, Computer Science and Biomedical Engineering, AGH University of Science and Technology, 30-059 Krakow, Poland
5 shared publications Andrzej Urbanik
Department of Radiology, Jagiellonian University Medical College, 19 Kopernika Street, 31-501 Krakow, Poland
4 shared publicationsKamil Krupa
Department of Radiology, Jagiellonian University Medical College, ul. Kopernika 19, 31-501 Krakow, Poland
3 shared publicationsSzymon Ciuk
Department of General and Interventional Radiology and Neuroradiology, Wrocław Medical University, Wrocław, Poland
3 shared publications
Latest Publications
- Open Access
- 12 December 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (24) , 11605
New Approaches to AI Methods for Screening Cardiomegaly on Chest Radiographs
Background: Cardiothoracic ratio (CTR) and transverse cardiac diameter (TCD) are parameters that are used to assess cardiac size on chest radiographs (CXRs). We aimed to investigate the performance and efficiency of artificial intelligence (AI) in screening for cardiomegaly on CXRs. Methods: The U-net architecture was designed for lung and heart segmentation. The CTR and TCD were then calculated using these labels and a mathematical algorithm. For the training set, we retrospectively included 65 randomly selected patients who underwent CXRs, while for the testing set, we chose 50 patients who underwent cardiac magnetic resonance (CMR) imaging and had available CXRs in the medical documentation. Results: Using U-net for the training set, the Dice coefficient for the lung was 0.984 ± 0.003 (min. 0.977), while for the heart it was 0.983 ± 0.004 (min. 0.972). For the testing set, the Dice coefficient for the lung was 0.970 ± 0.012 (min. 0.926), while for the heart it was 0.950 ± 0.021 (min. 0.871). The mean CTR and TCD measurements were slightly greater when calculated from either manual or automated segmentation than when manually read. Receiver operating characteristic analyses showed that both the CTR and TCD measurements calculated from either manual or automated segmentation, or when manually read, were good predictors of cardiomegaly diagnosed in CMR. However, McNemar tests have shown that diagnoses made with TCD, rather than CTR, were more consistent with CMR diagnoses. According to a different definition of cardiomegaly based on CMR imaging, accuracy for CTR measurements ranged from 62.0 to 74.0% for automatic segmentation (for TCD it ranged from 64.0 to 72.0%). Conclusion: The use of AI may optimize the screening process for cardiomegaly on CXRs. Future studies should focus on improving the accuracy of AI algorithms and on assessing the usefulness both of CTR and TCD measurements in screening for cardiomegaly.
- 12 December 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (24) , 11605
Open AccessNew Approaches to AI Methods for Screening Cardiomegaly on Chest Radiographs
Background: Cardiothoracic ratio (CTR) and transverse cardiac diameter (TCD) are parameters that are used to assess cardiac size on chest radiographs (CXRs). We aimed to investigate the performance and efficiency of artificial intelligence (AI) in screening for cardiomegaly on CXRs. Methods: The U-net architecture was designed for lung and heart segmentation. The CTR and TCD were then calculated using these labels and a mathematical algorithm. For the training set, we retrospectively included 65 randomly selected patients who underwent CXRs, while for the testing set, we chose 50 patients who underwent cardiac magnetic resonance (CMR) imaging and had available CXRs in the medical documentation. Results: Using U-net for the training set, the Dice coefficient for the lung was 0.984 ± 0.003 (min. 0.977), while for the heart it was 0.983 ± 0.004 (min. 0.972). For the testing set, the Dice coefficient for the lung was 0.970 ± 0.012 (min. 0.926), while for the heart it was 0.950 ± 0.021 (min. 0.871). The mean CTR and TCD measurements were slightly greater when calculated from either manual or automated segmentation than when manually read. Receiver operating characteristic analyses showed that both the CTR and TCD measurements calculated from either manual or automated segmentation, or when manually read, were good predictors of cardiomegaly diagnosed in CMR. However, McNemar tests have shown that diagnoses made with TCD, rather than CTR, were more consistent with CMR diagnoses. According to a different definition of cardiomegaly based on CMR imaging, accuracy for CTR measurements ranged from 62.0 to 74.0% for automatic segmentation (for TCD it ranged from 64.0 to 72.0%). Conclusion: The use of AI may optimize the screening process for cardiomegaly on CXRs. Future studies should focus on improving the accuracy of AI algorithms and on assessing the usefulness both of CTR and TCD measurements in screening for cardiomegaly.
- Open Access
- 12 March 2024
- Published by Elsevier in Medical Image Analysis
- Vol. 94, 103141
Use of superpixels for improvement of inter-rater and intra-rater reliability during annotation of medical images
- 12 March 2024
- Published by Elsevier in Medical Image Analysis
- Vol. 94, 103141
Open AccessUse of superpixels for improvement of inter-rater and intra-rater reliability during annotation of medical images
- Open Access
- 28 September 2023
- Published by IOS Press
- p. 1835-1842
CompLung: Comprehensive Computer-Aided Diagnosis of Lung Cancer
Lung cancer is a leading cause of cancer-related deaths, and early diagnosis is crucial for its effective treatment. That is why computer-aided tools have been developed to support particular steps of CT scan analysis, including lung segmentation, suspicious region detection, and patient-level diagnosis. However, none of the previous approaches addressed this process comprehensively. To fill this gap, we introduce CompLung, a comprehensive tool for lung cancer diagnosis that performs all of the above-listed steps in an end-to-end manner. We have trained the CompLung architecture using the publicly available LIDC-IDRI dataset extended with lung segmentation masks obtained from our internal radiologists, which we make publicly available to boost the research on this emerging topic. Finally, we conduct extensive experiments and demonstrate the superior performance and interpretability of CompLung compared to existing methods for lung cancer diagnosis.
- 28 September 2023
- Published by IOS Press
- p. 1835-1842
Open AccessCompLung: Comprehensive Computer-Aided Diagnosis of Lung Cancer
Lung cancer is a leading cause of cancer-related deaths, and early diagnosis is crucial for its effective treatment. That is why computer-aided tools have been developed to support particular steps of CT scan analysis, including lung segmentation, suspicious region detection, and patient-level diagnosis. However, none of the previous approaches addressed this process comprehensively. To fill this gap, we introduce CompLung, a comprehensive tool for lung cancer diagnosis that performs all of the above-listed steps in an end-to-end manner. We have trained the CompLung architecture using the publicly available LIDC-IDRI dataset extended with lung segmentation masks obtained from our internal radiologists, which we make publicly available to boost the research on this emerging topic. Finally, we conduct extensive experiments and demonstrate the superior performance and interpretability of CompLung compared to existing methods for lung cancer diagnosis.
- Open Access
Deep Learning Algorithm for Differentiating Patients with a Healthy Liver from Patients with Liver Lesions Based on MR Images
The problems in diagnosing the state of a vital organ such as the liver are complex and remain unresolved. These problems are underscored by frequently published studies on this issue. At the same time, demand for imaging diagnostics, preferably using a method that can detect the disease at the earliest possible stage, is constantly increasing. In this paper, we present liver diseases in the context of diagnosis, diagnostic problems, and possible elimination. We discuss the dataset and methods and present the stages of the pipeline we developed, leading to multiclass segmentation of the liver in multiparametric MR image into lesions and normal tissue. Finally, based on the processing results, each case is classified as either a healthy liver or a liver with lesions. For the training set, the AUC ROC is 0.925 (standard error 0.013 and a p-value less than 0.001), and for the test set, the AUC ROC is 0.852 (standard error 0.039 and a p-value less than 0.001). Further refinements to the proposed pipeline are also discussed. The proposed approach could be used in the detection of focal lesions in the liver and the description of liver tumors. Practical application of the developed multi-class segmentation method represents a key step toward standardizing the medical evaluation of focal lesions in the liver.
Open AccessDeep Learning Algorithm for Differentiating Patients with a Healthy Liver from Patients with Liver Lesions Based on MR Images
The problems in diagnosing the state of a vital organ such as the liver are complex and remain unresolved. These problems are underscored by frequently published studies on this issue. At the same time, demand for imaging diagnostics, preferably using a method that can detect the disease at the earliest possible stage, is constantly increasing. In this paper, we present liver diseases in the context of diagnosis, diagnostic problems, and possible elimination. We discuss the dataset and methods and present the stages of the pipeline we developed, leading to multiclass segmentation of the liver in multiparametric MR image into lesions and normal tissue. Finally, based on the processing results, each case is classified as either a healthy liver or a liver with lesions. For the training set, the AUC ROC is 0.925 (standard error 0.013 and a p-value less than 0.001), and for the test set, the AUC ROC is 0.852 (standard error 0.039 and a p-value less than 0.001). Further refinements to the proposed pipeline are also discussed. The proposed approach could be used in the detection of focal lesions in the liver and the description of liver tumors. Practical application of the developed multi-class segmentation method represents a key step toward standardizing the medical evaluation of focal lesions in the liver.
- Open Access
- 1 February 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (3) , 1882
What Changes Occur in the Brain of Veteran? A Magnetic Resonance Imaging and Proton Magnetic Resonance Spectroscopy Study
The aims of this study were to assess the common anomalies in the MRI examinations of the heads of soldiers as well as to compare the relative concentration of magnetic resonance spectroscopy (MRS) metabolites in the brains of soldiers with those of healthy age-matched controls. Overall, 54 professional male soldiers were included in the study group and 46 healthy, age-matched males were in the control group. The relative values of N-acetylaspartate (NAA), choline (Cho), and myoinositol (mI) to creatine (Cr) were assessed. The mean relative concentrations of metabolites were compared between the study and the control group, separately for the frontal and occipital lobes, as well as between the right and left hemispheres within the study group only. The most frequent findings in the head MRI of the soldiers were: asymmetric lateral ventricles and dilated perivascular spaces, enlargement of the subarachnoid spaces, and the presence of cavum septum pellucidum and cavum vergae; the high frequency of sinus disease should also be noted. In the frontal lobes, the mI/Cr ratio was significantly higher (p = 0.005), while the NAA/Cr ratio was lower (p = 0.001), in the group of soldiers (vs. the study group). In the occipital lobes, the NAA/Cr ratio was significantly lower (p = 0.005) in the military personnel and there was a tendency to a higher mI/Cr ratio in the soldiers’ occipital lobes (p = 0.056) (vs. the study group). Comparing the metabolites between the left and right hemispheres in soldiers preferring a right shooting position, a significantly higher mI/Cr (p < 0.001) ratio was observed in the right frontal lobe (vs. the left) and a markedly lower NAA/Cr (p = 0.003) in the right occipital lobe (vs. the left). These changes are associated with astrogliosis and neuronal loss, presumably secondary to repetitive mild traumatic brain injury.
- 1 February 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (3) , 1882
Open AccessWhat Changes Occur in the Brain of Veteran? A Magnetic Resonance Imaging and Proton Magnetic Resonance Spectroscopy Study
The aims of this study were to assess the common anomalies in the MRI examinations of the heads of soldiers as well as to compare the relative concentration of magnetic resonance spectroscopy (MRS) metabolites in the brains of soldiers with those of healthy age-matched controls. Overall, 54 professional male soldiers were included in the study group and 46 healthy, age-matched males were in the control group. The relative values of N-acetylaspartate (NAA), choline (Cho), and myoinositol (mI) to creatine (Cr) were assessed. The mean relative concentrations of metabolites were compared between the study and the control group, separately for the frontal and occipital lobes, as well as between the right and left hemispheres within the study group only. The most frequent findings in the head MRI of the soldiers were: asymmetric lateral ventricles and dilated perivascular spaces, enlargement of the subarachnoid spaces, and the presence of cavum septum pellucidum and cavum vergae; the high frequency of sinus disease should also be noted. In the frontal lobes, the mI/Cr ratio was significantly higher (p = 0.005), while the NAA/Cr ratio was lower (p = 0.001), in the group of soldiers (vs. the study group). In the occipital lobes, the NAA/Cr ratio was significantly lower (p = 0.005) in the military personnel and there was a tendency to a higher mI/Cr ratio in the soldiers’ occipital lobes (p = 0.056) (vs. the study group). Comparing the metabolites between the left and right hemispheres in soldiers preferring a right shooting position, a significantly higher mI/Cr (p < 0.001) ratio was observed in the right frontal lobe (vs. the left) and a markedly lower NAA/Cr (p = 0.003) in the right occipital lobe (vs. the left). These changes are associated with astrogliosis and neuronal loss, presumably secondary to repetitive mild traumatic brain injury.
- Open Access
- 6 June 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 41 (11) , 3231-3241
Benchmarking of Deep Architectures for Segmentation of Medical Images
In recent years, there were many suggestions regarding modifications of the well-known U-Net architecture in order to improve its performance. The central motivation of this work is to provide a fair comparison of U-Net and its five extensions using identical conditions to disentangle the influence of model architecture, model training, and parameter settings on the performance of a trained model. For this purpose each of these six segmentation architectures is trained on the same nine data sets. The data sets are selected to cover various imaging modalities (X-rays, computed tomography, magnetic resonance imaging), single- and multi-class segmentation problems, and single- and multi-modal inputs. During the training, it is ensured that the data preprocessing, data set split into training, validation, and testing subsets, optimizer, learning rate change strategy, architecture depth, loss function, supervision and inference are exactly the same for all the architectures compared. Performance is evaluated in terms of Dice coefficient, surface Dice coefficient, average surface distance, Hausdorff distance, training, and prediction time. The main contribution of this experimental study is demonstrating that the architecture variants do not improve the quality of inference related to the basic U-Net architecture while resource demand rises.
- 6 June 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 41 (11) , 3231-3241
Open AccessBenchmarking of Deep Architectures for Segmentation of Medical Images
In recent years, there were many suggestions regarding modifications of the well-known U-Net architecture in order to improve its performance. The central motivation of this work is to provide a fair comparison of U-Net and its five extensions using identical conditions to disentangle the influence of model architecture, model training, and parameter settings on the performance of a trained model. For this purpose each of these six segmentation architectures is trained on the same nine data sets. The data sets are selected to cover various imaging modalities (X-rays, computed tomography, magnetic resonance imaging), single- and multi-class segmentation problems, and single- and multi-modal inputs. During the training, it is ensured that the data preprocessing, data set split into training, validation, and testing subsets, optimizer, learning rate change strategy, architecture depth, loss function, supervision and inference are exactly the same for all the architectures compared. Performance is evaluated in terms of Dice coefficient, surface Dice coefficient, average surface distance, Hausdorff distance, training, and prediction time. The main contribution of this experimental study is demonstrating that the architecture variants do not improve the quality of inference related to the basic U-Net architecture while resource demand rises.
- 1 February 2021
- Published by Elsevier in Clinical Imaging
- Vol. 70, 81-88
Endovascular embolization of wide-necked bifurcation aneurysms with the use of pCONus device: A systematic review and meta-analysis
- 1 February 2021
- Published by Elsevier in Clinical Imaging
- Vol. 70, 81-88
Endovascular embolization of wide-necked bifurcation aneurysms with the use of pCONus device: A systematic review and meta-analysis
- 1 April 2020
- Published by Elsevier in Biocybernetics and Biomedical Engineering
- Vol. 40 (2) , 729-739
A fast graph-based algorithm for automated segmentation of subcutaneous and visceral adipose tissue in 3D abdominal computed tomography images
- 1 April 2020
- Published by Elsevier in Biocybernetics and Biomedical Engineering
- Vol. 40 (2) , 729-739
A fast graph-based algorithm for automated segmentation of subcutaneous and visceral adipose tissue in 3D abdominal computed tomography images
- Open Access
- 18 January 2020
- Published by Springer Nature in Rheumatology International
- Vol. 40 (4) , 625-633
The semi-automated algorithm for the detection of bone marrow oedema lesions in patients with axial spondyloarthritis
The aim of the study was to create the efficient tool for semi-automated detection of bone marrow oedema lesions in patients with axial spondyloarthritis (axSpA). MRI examinations of 22 sacroiliac joints of patients with confirmed axSpA-related sacroiliitis (median SPARCC score: 14 points) were included into the study. Design of our algorithm is based on Maksymowych et al. evaluation method and consists of the following steps: manual segmentation of bones (T1W sequence), automated detection of reference signal region, sacroiliac joint central lines and ROIs, a division of ROIs into quadrants, automated detection of inflammatory changes (STIR sequence). As a gold standard, two sets of manual lesion delineations were created. Two approaches to the performance assessment of lesion detection were considered: pixel-wise (detections compared pixel by pixel) and quadrant-wise (quadrant to quadrant). Statistical analysis was performed using Spearman's correlation coefficient. Correlation coefficient obtained for pixel-wise comparison of semi-automated and manual detections was 0.87 (p = 0.001), while for quadrant-wise analysis was 0.83 (p = 0.001). The correlation between two sets of manual detections was 0.91 for pixel-wise comparison (p = 0.001) and 0.88 (p = 0.001) for quadrant-wise approach. Spearman’s correlation between two manual assessments was not statistically different from the correlation between semi-automated and manual evaluations, both for pixel- (p = 0.14) and quadrant-wise (p = 0.17) analysis. Average single slice processing time: 0.64 ± 0.30 s. Our method allows for objective detection of bone marrow oedema lesions in patients with axSpA. The quantification of affected pixels and quadrants has comparable reliability to manual assessment.
- 18 January 2020
- Published by Springer Nature in Rheumatology International
- Vol. 40 (4) , 625-633
Open AccessThe semi-automated algorithm for the detection of bone marrow oedema lesions in patients with axial spondyloarthritis
The aim of the study was to create the efficient tool for semi-automated detection of bone marrow oedema lesions in patients with axial spondyloarthritis (axSpA). MRI examinations of 22 sacroiliac joints of patients with confirmed axSpA-related sacroiliitis (median SPARCC score: 14 points) were included into the study. Design of our algorithm is based on Maksymowych et al. evaluation method and consists of the following steps: manual segmentation of bones (T1W sequence), automated detection of reference signal region, sacroiliac joint central lines and ROIs, a division of ROIs into quadrants, automated detection of inflammatory changes (STIR sequence). As a gold standard, two sets of manual lesion delineations were created. Two approaches to the performance assessment of lesion detection were considered: pixel-wise (detections compared pixel by pixel) and quadrant-wise (quadrant to quadrant). Statistical analysis was performed using Spearman's correlation coefficient. Correlation coefficient obtained for pixel-wise comparison of semi-automated and manual detections was 0.87 (p = 0.001), while for quadrant-wise analysis was 0.83 (p = 0.001). The correlation between two sets of manual detections was 0.91 for pixel-wise comparison (p = 0.001) and 0.88 (p = 0.001) for quadrant-wise approach. Spearman’s correlation between two manual assessments was not statistically different from the correlation between semi-automated and manual evaluations, both for pixel- (p = 0.14) and quadrant-wise (p = 0.17) analysis. Average single slice processing time: 0.64 ± 0.30 s. Our method allows for objective detection of bone marrow oedema lesions in patients with axSpA. The quantification of affected pixels and quadrants has comparable reliability to manual assessment.
- Open Access
- 10 July 2019
- Published by Springer Nature in Rheumatology International
- Vol. 39 (9) , 1559-1565
The usefulness of diffusion-weighted imaging (DWI) and dynamic contrast-enhanced (DCE) sequences visual assessment in the early diagnosis of axial spondyloarthritis
The aim of the study was to compare the diagnostic efficacy of the visual assessment of diffusion-weighted imaging (DWI) and dynamic contrast-enhanced (DCE) sequences compared to the STIR sequence in the diagnostics of active sacroiliitis in the course of axial spondyloarthritis (axSpA). The study group consisted of 49 patients who had undergone multiparametric magnetic resonance imaging of the sacroiliac joints (SIJs) due to clinical suspicion of axSpA. Two independent observers retrospectively assessed four quadrants of the SIJs for the presence of subchondral bone marrow oedema/osteitis with the use of modified SPARCC score in sequences: STIR, DWI (with ADC map) and DCE. Diagnostic efficiency parameters were calculated for DWI and DCE sequence separately, using STIR sequence as a reference. Inter-observer agreement was evaluated with the use of kappa coefficient. Patients' clinical symptoms were analysed to identify the group fulfilling the imaging arm of the ASAS criteria for axSpA. Overall, 46.9% (n = 23) of patients fulfilled the imaging arm of ASAS criteria for axial spondyloarthritis. DWI with ADC map: accuracy 95.6%, sensitivity 99.4%, specificity 54.0%. DCE sequence: accuracy 96.8%, sensitivity 98.4%, specificity 79.5%. The highest level of inter-observer agreement was achieved for STIR sequence (kappa = 0.888), slightly lower for DCE sequence (kappa = 0.773) and the lowest for DWI with ADC (kappa = 0.674). Visual assessment of the DWI and DCE sequences has high accuracy and sensitivity of bone marrow oedema/osteitis detection, but the specificity and inter-observer agreement are poor, especially for the DWI sequence with ADC maps.
- 10 July 2019
- Published by Springer Nature in Rheumatology International
- Vol. 39 (9) , 1559-1565
Open AccessThe usefulness of diffusion-weighted imaging (DWI) and dynamic contrast-enhanced (DCE) sequences visual assessment in the early diagnosis of axial spondyloarthritis
The aim of the study was to compare the diagnostic efficacy of the visual assessment of diffusion-weighted imaging (DWI) and dynamic contrast-enhanced (DCE) sequences compared to the STIR sequence in the diagnostics of active sacroiliitis in the course of axial spondyloarthritis (axSpA). The study group consisted of 49 patients who had undergone multiparametric magnetic resonance imaging of the sacroiliac joints (SIJs) due to clinical suspicion of axSpA. Two independent observers retrospectively assessed four quadrants of the SIJs for the presence of subchondral bone marrow oedema/osteitis with the use of modified SPARCC score in sequences: STIR, DWI (with ADC map) and DCE. Diagnostic efficiency parameters were calculated for DWI and DCE sequence separately, using STIR sequence as a reference. Inter-observer agreement was evaluated with the use of kappa coefficient. Patients' clinical symptoms were analysed to identify the group fulfilling the imaging arm of the ASAS criteria for axSpA. Overall, 46.9% (n = 23) of patients fulfilled the imaging arm of ASAS criteria for axial spondyloarthritis. DWI with ADC map: accuracy 95.6%, sensitivity 99.4%, specificity 54.0%. DCE sequence: accuracy 96.8%, sensitivity 98.4%, specificity 79.5%. The highest level of inter-observer agreement was achieved for STIR sequence (kappa = 0.888), slightly lower for DCE sequence (kappa = 0.773) and the lowest for DWI with ADC (kappa = 0.674). Visual assessment of the DWI and DCE sequences has high accuracy and sensitivity of bone marrow oedema/osteitis detection, but the specificity and inter-observer agreement are poor, especially for the DWI sequence with ADC maps.