Engineering design and research

2 reviewers · Major revision · revised 18 July 2026 · accepted 31 July 2026

Both reviewers returned major revision. The resubmission expanded the recommendation engine's scoring in Methods, added Discussion paragraphs on error propagation and patient safety and on the ethical, regulatory and medico-legal position, widened the comparison with published prosthodontics AI, and restructured Future Work around clinical and usability evaluation.

2,990 words · 240-word structured abstract · 17 Vancouver references

YOLOv8-Nano (~3.01M parameters), fine-tuned from COCO-pretrained yolov8n.pt on 16,995 panoramic radiographs across 10 condition classes. Dental X-Ray Panoramic v3, CC BY 4.0.

The paper is fixed and reports the validation split it was written against; the test row is the shipped checkpoint on images it never saw, recorded with its weights digest.

• No clinical validation. The engine's cost ranges, age windows and severity weights are researcher-selected heuristics, not calibrated against expert consensus.
• The paper's figures are validation-split, not held-out; the test-split evaluation came after submission.
• The knowledge base encodes Sri Lankan Rupee costs and local clinical convention, so another healthcare system would need a local review before use.
• The output is advisory. Clinical accountability stays entirely with the treating clinician, and deployment beyond research would need software-as-a-medical-device conformity in the target jurisdiction.