Philipp Nikolaus Maintz
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Philipp Nikolaus Maintz has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Bernhard Schmidt
Siemens Healthcare GmbH, 91052 Erlangen, Germany
1 shared publication - Cagan Koska
Department of Electrical Electronical, Engineering Faculty, Yaşar University, 35100 Izmir, Türkiye
1 shared publication - Falko Ensle
Department of Radiology and Imaging, Hospital for Special Surgery, 535 E 70th St, New York, NY 10021
1 shared publication - Hatem Alkadhi
Diagnostic and Interventional Radiology, University Hospital Zurich, University of Zurich, Zurich, Switzerland
1 shared publication - Hubertus Pietsch
From the Bayer AG, Radiology, Berlin, Germany (T.F., H.P., L.S., J.E.); Wels - Omnino Medico, Rotkreuz, Switzerland (T.W.); Bayer U.S. LLC, Indianola, PA (P.S.); and Department of Gynecology, Obstetrics and Reproductive Medicine, University Medical School of Saarland, Homburg/Saar, Germany (J.E.)
1 shared publication - Ilker Ozgur Koska
Department of Radiology, Behçet Uz Children’s Hospital, Izmir, Turkey
1 shared publication
Latest Publications
- Open Access
- 31 July 2026
- Published by MDPI AG in Diagnostics
- Vol. 16 (15) , 2425
Quantifying User Satisfaction: Weighted Metric Approach for Evaluating Deep Learning-Based Thigh MRI Segmentations
Background: The aim of this paper is to develop a clinically useful benchmark signature of deep learning (DL)-based MRI segmentations and objectively predict user satisfaction based on a weighted combination of multiple performance metrics. Methods: This post hoc analysis of a prospective study analyzed MRI data from 68 patients (29.4 ± 5.6 years; 45% female) acquired during a randomized clinical trial. Fat fraction maps of axial Dixon MRI were used to segment six classes of the thigh. Two radiologists qualitatively scored DL-based segmentations using a 5-point Likert scale. A hybrid deep learning model (MobileNetV2 and DINOv2) was developed to predict Likert scores directly from MRI slices and segmentation maps. Additionally, a weighted combination of quantitative metrics (Dice score, Hausdorff distance, Jaccard index) was determined to best correlate with the Likert scores. Statistical analyses included Cohen’s kappa, Spearman correlation, and regression model performance (MAE, RMSE, ROC_AUC). Results: Interreader agreement for Likert scores was near-perfect (κ = 0.82). The DINOv2 model achieved lower MAE (0.2343) and RMSE (0.1129) than MobileNetV2 (0.3488 and 0.431, respectively) in predicting Likert scores. The weighted metric combination model, using Lasso regression, predicted Likert = 5 with 84.1% accuracy, 100% recall, and ROC_AUC of 0.824. Individual metrics showed weak-to-moderate correlation with Likert scores, with Dice and Jaccard indices for extensor intramuscular fat exhibiting the highest correlation (r = 0.37 and r = 0.34, respectively). Conclusions: The proposed DL-based model accurately predicted radiologist-assigned Likert five scores, while the weighted metric combination model aligned closely with user satisfaction, offering a practical alternative to manual validation for clinical adoption of automated segmentation tools.
- 31 July 2026
- Published by MDPI AG in Diagnostics
- Vol. 16 (15) , 2425
Open AccessQuantifying User Satisfaction: Weighted Metric Approach for Evaluating Deep Learning-Based Thigh MRI Segmentations
Background: The aim of this paper is to develop a clinically useful benchmark signature of deep learning (DL)-based MRI segmentations and objectively predict user satisfaction based on a weighted combination of multiple performance metrics. Methods: This post hoc analysis of a prospective study analyzed MRI data from 68 patients (29.4 ± 5.6 years; 45% female) acquired during a randomized clinical trial. Fat fraction maps of axial Dixon MRI were used to segment six classes of the thigh. Two radiologists qualitatively scored DL-based segmentations using a 5-point Likert scale. A hybrid deep learning model (MobileNetV2 and DINOv2) was developed to predict Likert scores directly from MRI slices and segmentation maps. Additionally, a weighted combination of quantitative metrics (Dice score, Hausdorff distance, Jaccard index) was determined to best correlate with the Likert scores. Statistical analyses included Cohen’s kappa, Spearman correlation, and regression model performance (MAE, RMSE, ROC_AUC). Results: Interreader agreement for Likert scores was near-perfect (κ = 0.82). The DINOv2 model achieved lower MAE (0.2343) and RMSE (0.1129) than MobileNetV2 (0.3488 and 0.431, respectively) in predicting Likert scores. The weighted metric combination model, using Lasso regression, predicted Likert = 5 with 84.1% accuracy, 100% recall, and ROC_AUC of 0.824. Individual metrics showed weak-to-moderate correlation with Likert scores, with Dice and Jaccard indices for extensor intramuscular fat exhibiting the highest correlation (r = 0.37 and r = 0.34, respectively). Conclusions: The proposed DL-based model accurately predicted radiologist-assigned Likert five scores, while the weighted metric combination model aligned closely with user satisfaction, offering a practical alternative to manual validation for clinical adoption of automated segmentation tools.
- Open Access
- 1 December 2025
- Published by Wolters Kluwer Health in Investigative Radiology
- Vol. 61 (8) , 568-574
High-Z Contrast Media for Coronary Photon-counting Detector CT Angiography
Background: Blooming artifacts from calcified plaques can obscure the vessel lumen, leading to overestimation of stenosis severity. Spectral coronary angiography with photon-counting detector CT (PCD-CT) provides virtual monoenergetic images (VMIs) for coronary artery disease assessment. While VMIs at high VMI energy levels reduce calcium blooming, iodine contrast is diminished, limiting diagnostic value. This study evaluated whether contrast media with an atomic number higher than iodine (high-Z) preserve vascular contrast using high VMI energy levels, thereby improving the accuracy of stenosis quantification. Methods: A phantom with 4 and 6 mm diameter rods to mimic small diameter vessels containing eccentric calcified plaques causing 25%, 50%, and 75% diameter stenoses was scanned with a dual-source PCD-CT system. Five different contrast media, including iodine, tungsten, holmium, hafnium, and bismuth, were tested. VMIs were reconstructed from 40 to 190 keV in 1-keV steps. Vessel attenuation, contrast-to-noise ratio (CNR), and stenoses were measured. Qualitative assessment of image quality was performed. Results: Iodine attenuation was high at lower VMI energy levels and dropped below 250 HU at >100 keV. Tungsten, holmium, hafnium, and bismuth maintained >250 HU attenuation throughout the entire energy range. Vessel CNR of iodine was high at lower and decreased at higher VMI energy levels, similar to the CNR of holmium and bismuth, though to a lesser extent. In distinction, CNRs of tungsten and hafnium were lower at lower VMI energy levels and increased to a relatively constant level at higher keV. Tungsten CNR increased with energy, approaching ~40 at high keV. Across all contrast media and stenosis degrees, stenoses were overestimated on low VMI energy levels (24% to 32.5% at 40 keV), while the degree of overestimation decreased at higher VMI energy levels (0% to 13.5% at 190 keV). At 190 keV, tungsten, hafnium, and bismuth showed ≤2.5% stenosis overestimation, compared with iodine (10% to 13.5%). Image quality varied between contrast media and energy levels: new very high-Z contrast media achieved higher scores, while iodine peaked at lower keV (55 to 70 keV) and, due to loss of contrast at higher energies, received the lowest overall scores. Conclusions: As compared with iodine, very high-Z contrast media enable superior lumen definition and more accurate stenosis assessment, also at high VMI energy levels, which minimize calcium blooming.
- 1 December 2025
- Published by Wolters Kluwer Health in Investigative Radiology
- Vol. 61 (8) , 568-574
Open AccessHigh-Z Contrast Media for Coronary Photon-counting Detector CT Angiography
Background: Blooming artifacts from calcified plaques can obscure the vessel lumen, leading to overestimation of stenosis severity. Spectral coronary angiography with photon-counting detector CT (PCD-CT) provides virtual monoenergetic images (VMIs) for coronary artery disease assessment. While VMIs at high VMI energy levels reduce calcium blooming, iodine contrast is diminished, limiting diagnostic value. This study evaluated whether contrast media with an atomic number higher than iodine (high-Z) preserve vascular contrast using high VMI energy levels, thereby improving the accuracy of stenosis quantification. Methods: A phantom with 4 and 6 mm diameter rods to mimic small diameter vessels containing eccentric calcified plaques causing 25%, 50%, and 75% diameter stenoses was scanned with a dual-source PCD-CT system. Five different contrast media, including iodine, tungsten, holmium, hafnium, and bismuth, were tested. VMIs were reconstructed from 40 to 190 keV in 1-keV steps. Vessel attenuation, contrast-to-noise ratio (CNR), and stenoses were measured. Qualitative assessment of image quality was performed. Results: Iodine attenuation was high at lower VMI energy levels and dropped below 250 HU at >100 keV. Tungsten, holmium, hafnium, and bismuth maintained >250 HU attenuation throughout the entire energy range. Vessel CNR of iodine was high at lower and decreased at higher VMI energy levels, similar to the CNR of holmium and bismuth, though to a lesser extent. In distinction, CNRs of tungsten and hafnium were lower at lower VMI energy levels and increased to a relatively constant level at higher keV. Tungsten CNR increased with energy, approaching ~40 at high keV. Across all contrast media and stenosis degrees, stenoses were overestimated on low VMI energy levels (24% to 32.5% at 40 keV), while the degree of overestimation decreased at higher VMI energy levels (0% to 13.5% at 190 keV). At 190 keV, tungsten, hafnium, and bismuth showed ≤2.5% stenosis overestimation, compared with iodine (10% to 13.5%). Image quality varied between contrast media and energy levels: new very high-Z contrast media achieved higher scores, while iodine peaked at lower keV (55 to 70 keV) and, due to loss of contrast at higher energies, received the lowest overall scores. Conclusions: As compared with iodine, very high-Z contrast media enable superior lumen definition and more accurate stenosis assessment, also at high VMI energy levels, which minimize calcium blooming.