Vollständiger Abstract
Worum geht es in dieser Arbeit?
High aggregate performance does not guarantee reliable individual predictions or clinical transportability. This study designed, implemented, and internally evaluated NEUROCALIB, a selective bimodular research framework comprising independently assessed classification and segmentation pipelines, and audited their integration into the NeuroVision web prototype. EfficientNetB0 was evaluated on 6,683 public brain MRI images across four classes; MobileNetV2–U-Net was evaluated on 4,048 image-mask pairs after removal of 189 exact duplicates. Validation data were used for model selection, temperature scaling, and abstention thresholds before the test sets were opened.
Bibliografischer Nachweis
Publikationsdaten
- Autor:innen
- Hervé Kinkete Mfumabi, Matangila Gradi, Nsiala Ndona Clémence, Kavugho Kahasa Thérèse, Kalema Manzambie Jordanie, Kema Mulumba Joël, Madio Motemona Godard, Ilolo Nana
- Quelle
- International Journal of Innovative Science and Research Technology
- Publikation
- 2026-01-01
- Band / Ausgabe
- Nicht angegeben
- Seiten
- Nicht angegeben
- ISSN / ISBN
- 2456-2165
- Zitationen
- 0 laut Crossref
- Referenzen
- 0 hinterlegt
Zitieren
Zitierfähiger Nachweis
Hervé Kinkete Mfumabi, Matangila Gradi, Nsiala Ndona Clémence, Kavugho Kahasa Thérèse, Kalema Manzambie Jordanie, Kema Mulumba Joël, Madio Motemona Godard, Ilolo Nana (2026). NEUROCALIB: Design, Implementation and Internal Evaluation of a Calibrated, Explainable and Selective Bimodular Framework for Brain Tumor Classification and Segmentation on MRI. International Journal of Innovative Science and Research Technology. https://doi.org/10.38124/ijisrt/26aug907