Johan Andreas Balle Rubak

Mr. Johan Andreas Balle Rubak

orcid

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

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

Representative Publications

View All Publications
16 January 2026 in Dentomaxillofacial Radiology
Verified
Journal Article
Closed access
JRJohan Andreas Balle Rubak
Johan Andreas Balle Rubak
Khuram NaveedKhuram Naveed
Khuram Naveed
SJSanyam Jain
Sanyam Jain
LELukas Esterle
Lukas Esterle
AIAlexandros Iosifidis
Alexandros Iosifidis
Ruben PauwelsRuben Pauwels
Ruben Pauwels
Objectives: Federated learning (FL) may mitigate privacy constraints, heterogeneous data quality, and inconsistent labelling in dental artificial intelligence (AI). FL was compared with centralized (CL) and local learning (LL) for tooth segmentation in panoramic radiographs across multiple data corruption scenarios. Methods: An Attention U-Net was trained on 2066 radiographs from 6 institutions across 4 settings: baseline (unaltered data), label manipulation (dilated/missing annotations), image-quality manipulation (additive Gaussian noise), and exclusion of one faulty client with corrupted data. FL was implemented via the Flower AI framework. Per-client training and validation loss trajectories were monitored for anomaly detection and a set of metrics (Dice, IoU, HD, HD95, and ASSD) were evaluated on a hold-out test set. From these metrics significance results were reported through Wilcoxon signed-rank test. CL and LL served as comparators. Results: Baseline: FL achieved a median Dice of 0.949 (ASSD: 1.332), slightly better than CL at 0.947 (ASSD: 1.371) and LL at 0.936-0.940 (ASSD: 1.519-1.698). Label manipulation: FL maintained the best median Dice score at 0.949 (ASSD: 1.465) versus CL’s 0.942 (ASSD: 1.757) and LL’s 0.930-0.940 (ASSD: 1.519-2.115). Similar performance was observed when 2 faulty clients were introduced. Image noise: FL led with a Dice at 0.949 (ASSD: 1.311); CL had a Dice of 0.948 (ASSD: 1.361); LL ranged from 0.932 to 0.940 (ASSD: 1.519-1.774). Similar performance was observed when 2 faulty clients were introduced, with CL performing slightly better than FL. Faulty client exclusion: FL showed a Dice of 0.948 (ASSD: 1.331) better than CL’s 0.946 (ASSD: 1.393). Loss curve monitoring reliably flagged the corrupted site. Conclusions: FL matches or exceeds CL and outperforms LL across corruption scenarios while preserving privacy. Per-client loss trajectories provide an effective anomaly-detection mechanism and support FL as a practical, privacy-preserving approach for scalable clinical AI development.

17 June 2026 in Applied Sciences
Verified
Journal Article
Open access
PDF file
JRJohan Andreas Balle Rubak
Johan Andreas Balle Rubak
SHSara Haghighat
Sara Haghighat
SJSanyam Jain
Sanyam Jain
Mostafa AldesokiMostafa Aldesoki
Mostafa Aldesoki
Akhilanand ChaurasiaAkhilanand Chaurasia
Akhilanand Chaurasia
Ruben PauwelsRuben Pauwels
Ruben Pauwels
Impaction of the mandibular third molar in proximity to the mandibular canal increases the risk of inferior alveolar nerve injury. Panoramic radiography is routinely used to assess this relationship. Automated classification of molar–canal overlap could support clinical triage and reduce unnecessary CBCT referrals, while Federated Learning (FL) enables multi-center collaboration without sharing patient data. We compared Local Learning (LL), FL, and Centralized Learning (CL) for binary overlap/no-overlap classification on cropped panoramic radiographs partitioned across eight independent labelers in a simulated heterogeneous multi-center setting. A pretrained ResNet-34 was trained under each paradigm and evaluated using per-client metrics with locally optimized thresholds and pooled test performance with a global threshold. Performance was assessed using area under the receiver operating characteristic curve (AUC) and threshold-based metrics, alongside training dynamics, Grad-CAM visualizations, and server-side aggregate monitoring signals. On the test set, CL achieved the highest performance (AUC 0.831; accuracy ≈ 0.782), FL showed intermediate performance (AUC 0.757; accuracy ≈ 0.703), and LL generalized poorly across clients (AUC range ≈ 0.619–0.734; mean ≈ 0.672). Training curves suggested overfitting, particularly in LL models, and Grad-CAM indicated more anatomically focused attention in CL and FL. Overall, centralized training provided the strongest performance, while FL offers a privacy-preserving alternative that outperforms LL.