Konstantinos Koukoutegos

Mr. Konstantinos Koukoutegos

University Hospitals Leuven, Campus Gasthuisberg

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
9 October 2024 in European Radiology Experimental
Verified
Journal Article
Open access
PDF file
KKKonstantinos Koukoutegos
Konstantinos Koukoutegos
RHRichard ’s Heeren
Richard ’s Heeren
LWLiesbeth De Wever
Liesbeth De Wever
FKFrederik De Keyzer
Frederik De Keyzer
FMFrederik Maes
Frederik Maes
HBHilde Bosmans
Hilde Bosmans
Background: Renal quantitative measurements are important descriptors for assessing kidney function. We developed a deep learning-based method for automated kidney measurements from computed tomography (CT) images. Methods: The study datasets comprised potential kidney donors (n = 88), both contrast-enhanced (Dataset 1 CE) and noncontrast (Dataset 1 NC) CT scans, and test sets of contrast-enhanced cases (Test set 2, n = 18), cases from a photon-counting (PC)CT scanner reconstructed at 60 and 190 keV (Test set 3 PCCT, n = 15), and low-dose cases (Test set 4, n = 8), which were retrospectively analyzed to train, validate, and test two networks for kidney segmentation and subsequent measurements. Segmentation performance was evaluated using the Dice similarity coefficient (DSC). The quantitative measurements’ effectiveness was compared to manual annotations using the intraclass correlation coefficient (ICC). Results: The contrast-enhanced and noncontrast models demonstrated excellent reliability in renal segmentation with DSC of 0.95 (Test set 1 CE), 0.94 (Test set 2), 0.92 (Test set 3 PCCT) and 0.94 (Test set 1 NC), 0.92 (Test set 3 PCCT), and 0.93 (Test set 4). Volume estimation was accurate with mean volume errors of 4%, 3%, 6% mL (contrast test sets) and 4%, 5%, 7% mL (noncontrast test sets). Renal axes measurements (length, width, and thickness) had ICC values greater than 0.90 (p < 0.001) for all test sets, supported by narrow 95% confidence intervals. Conclusion: Two deep learning networks were shown to derive quantitative measurements from contrast-enhanced and noncontrast renal CT imaging at the human performance level. Relevance statement: Deep learning-based networks can automatically obtain renal clinical descriptors from both noncontrast and contrast-enhanced CT images. When healthy subjects comprise the training cohort, careful consideration is required during model adaptation, especially in scenarios involving unhealthy kidneys. This creates an opportunity for improved clinical decision-making without labor-intensive manual effort. Key Points: Trained 3D UNet models quantify renal measurements from contrast and noncontrast CT. The models performed interchangeably to the manual annotator and to each other. The models can provide expert-level, quantitative, accurate, and rapid renal measurements. Graphical

27 September 2021 in Methods and Protocols
Verified
Journal Article
Open access
PDF file
ADAlexandros Dimopoulos
Alexandros Dimopoulos
KKKonstantinos Koukoutegos
Konstantinos Koukoutegos
Fotis PsomopoulosFotis Psomopoulos
Fotis Psomopoulos
Panagiotis MoulosPanagiotis Moulos
Panagiotis Moulos
RNA sequencing has become the standard technique for high resolution genome-wide monitoring of gene expression. As such, it often comprises the first step towards understanding complex molecular mechanisms driving various phenotypes, spanning organ development to disease genesis, monitoring and progression. An advantage of RNA sequencing is its ability to capture complex transcriptomic events such as alternative splicing which results in alternate isoform abundance. At the same time, this advantage remains algorithmically and computationally challenging, especially with the emergence of even higher resolution technologies such as single-cell RNA sequencing. Although several algorithms have been proposed for the effective detection of differential isoform expression from RNA-Seq data, no widely accepted golden standards have been established. This fact is further compounded by the significant differences in the output of different algorithms when applied on the same data. In addition, many of the proposed algorithms remain scarce and poorly maintained. Driven by these challenges, we developed a novel integrative approach that effectively combines the most widely used algorithms for differential transcript and isoform analysis using state-of-the-art machine learning techniques. We demonstrate its usability by applying it on simulated data based on several organisms, and using several performance metrics; we conclude that our strategy outperforms the application of the individual algorithms. Finally, our approach is implemented as an R Shiny application, with the underlying data analysis pipelines also available as docker containers.

7 March 2024
Verified
Book Chapter
Closed access
PDF file
KKKonstantinos Koukoutegos
Konstantinos Koukoutegos
FMFrederik Maes
Frederik Maes
HBHilde Bosmans
Hilde Bosmans

12 November 2026 in European Radiology Experimental
Verified
Journal Article
Open access
PDF file
KKKonstantinos Koukoutegos
Konstantinos Koukoutegos
TOTom Oyen
Tom Oyen
RORaymond Oyen
Raymond Oyen
FKFrederik De Keyzer
Frederik De Keyzer
FMFrederik Maes
Frederik Maes
HBHilde Bosmans
Hilde Bosmans
Objectives: Accurate quantification of total kidney volume (TKV) and Mayo imaging classification (MIC) is essential for assessing disease severity in autosomal dominant polycystic kidney disease (ADPKD). This study aimed to develop and validate a deep learning–based method using noncontrast computed tomography (CT) scans to automate TKV measurement and MIC assignment. Materials and methods: A retrospective dataset of 60 ADPKD patients was used to train a three-dimensional (3D) UNet model for kidney segmentation and TKV estimation. Kidney axes were calculated using principal component analysis (PCA). The model was tested on consecutive scans from 30 patients (median interval: 3.6 years) and 28 external patient scans. Segmentation accuracy was evaluated using the Dice similarity coefficient (DSC), and measurement reliability was assessed using the intraclass correlation coefficient (ICC). MIC was assigned and compared across timepoints. Results: The model achieved high segmentation accuracy (DSC: 0.93 ± 0.04 [mean ± standard deviation] and 0.94 ± 0.03 for consecutive scans; 0.96 ± 0.08 for external validation). Manual and automated TKV measurements were strongly correlated (ICC: 0.94–0.99, p < 0.001), with a mean volume error of 7.61 ± 7.26%. MIC assignment matched in 90–93.3% of cases. Kidney dimensions (length > 16.5 cm, width > 9.5 cm, depth > 8 cm) predicted high-risk MIC classes with positive predictive values > 88%. Longitudinal analysis showed significant increases in TKV and declines in estimated glomerular filtration rate–eGFR ( p < 0.001). Conclusion: The proposed method enables accurate TKV quantification and MIC classification from noncontrast CT, supporting severity assessment and longitudinal monitoring in ADPKD. Key Points: Question Can a deep learning method using noncontrast CT accurately automate total kidney volume measurement and Mayo Imaging Classification assignment in autosomal dominant polycystic kidney disease? Findings In internal and external testing, automated kidney segmentation showed high agreement with reference measurements and correctly assigned Mayo Imaging Classification in most cases with overall accuracy. Relevance statement Automated analysis of noncontrast CT provided reliable kidney volume quantification and disease severity classification, reducing manual workload while enabling longitudinal assessment of disease progression in autosomal dominant polycystic kidney disease.

24 September 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
VPVaralakshmi Perumal
Varalakshmi Perumal
ALAkis Linardos
Akis Linardos
KKKonstantinos Koukoutegos
Konstantinos Koukoutegos
WTWenyi Tang
Wenyi Tang
EGEleftherios Garyfallidis
Eleftherios Garyfallidis
Spyridon BakasSpyridon Bakas
Spyridon Bakas
Federated learning (FL) has emerged as a promising novel paradigm for enabling multi-institutional collaborations and expediting data access for training and benchmarking Healthcare AI models, without the need for data centralization. During FL, each collaborator performs local computations on their own data and sends model updates to a central server, which collects and aggregates all model updates into a global consensus model. Several FL aggregation (FLAg) strategies have been developed, each focusing on improving the effectiveness of FL, while addressing distinct technical and data-related challenges – from handling varying resource environments, to patient privacy, and to data imbalance across collaborators. In this article, we provide a systematic survey and taxonomy of FLAg strategies designed for healthcare, paired with their relation to known FL challenges, in our attempt to guide readers navigate and interpret the current scattered body of relevant literature. Our analysis reveals that significant progress has been made in improving model utility and privacy guarantees, while mitigating stragglers, system heterogeneity, fairness and biases, uncertainty, security, and data imbalance. We also touch upon current research gaps between current methodological advances and real-world deployment in healthcare, including limitations in i) evaluation beyond model utility, ii) clinical risk assessment, and iii) governance requirements for transparent and accountable FL.