We are pleased to announce that our latest research has been published in IEEE Access (Open Access).
The study, titled “From Geometric to Realistic: A Pipelined Deep Learning Framework for Cranial Implant Design using PCA-Based Synthetic Data,” addresses the shortage of high-quality labeled data for automated cranial implant design. A Principal Component Analysis (PCA)-based generator produces realistic synthetic cranial defects, which are combined with clinical data to train deep networks for volumetric skull completion.
The proposed pipeline, which chains a boundary-specialized model with a volume-specialized model, achieves a Hausdorff Distance of 7.90 mm and an Implant DICE score of 81.89%, challenging the assumption that conventional geometric augmentation is sufficient for capturing the complexity of real defects. The work lays a foundation for fully autonomous, 3D-printable cranial implant design.
Authors:
Görkem Serbes (Yıldız Technical University)
Hamza Osman İlhan (Yıldız Technical University)
Osman Liv (TÜBİTAK)
Batuhan Tanrıverdi (Arçelik A.Ş.)
Fatih Ekrem Onat (Yıldız Technical University)
Yiğit Kahraman (Yıldız Technical University)
Berk Arıdağ (Yıldız Technical University)
Melih Fırat Budak (Yıldız Technical University)
Berke Cansız (Yıldız Technical University)
Berke Apaydınlı (Yıldız Technical University)
Abdulkadir Günay (Yıldız Technical University)
Yunus Emre Çakmaklı (Yıldız Technical University)
Nuri Serdar Baş (University of Health Sciences, Kanuni Sultan Süleyman Training and Research Hospital)
Mihrigül Ekşi Altan (Yıldız Technical University)
