Deep Learning-Based Automated Anatomical Landmark Detection and Saw Blade Size Prediction for Canine Tibial Plateau...

Area of Science:

• Veterinary orthopedics
• Medical imaging analysis
• Artificial intelligence in veterinary medicine

Background:

• Tibial plateau leveling osteotomy (TPLO) is a common surgical procedure for canine cranial cruciate ligament rupture.
• Accurate preoperative planning, including tibial plateau angle (TPA) measurement, is crucial for successful TPLO outcomes.
• Current methods for TPA measurement can be time-consuming and subject to inter-observer variability.

Purpose of the Study:

• To develop and validate a fully automated deep learning workflow for canine hindlimb lateral radiographs.
• To localize key anatomical landmarks, derive the tibial plateau angle (TPA), and recommend saw blade size for TPLO preoperative planning.
• To assess the accuracy and clinical utility of the automated system.

Main Methods:

• A retrospective validation study utilizing 200 annotated lateral radiographs from 130 dogs.
• A custom four-stage U-Net model trained on multiple image representations to detect TPLO-related landmarks.
• A deterministic geometric module for TPA calculation and saw blade size recommendation.

Main Results:

• The automated system achieved a mean absolute error of 1.34 ± 1.73° for TPA prediction, with 82% of cases within 2° of surgeon reference.
• Saw blade size prediction showed a mean absolute error of 0.32 ± 0.85 mm, with 87.5% exact agreement.
• High correlation (Pearson's r=0.87) and acceptable limits of agreement for TPA were observed.

Conclusions:

• The developed deep learning workflow offers clinically useful automated TPA and saw blade size estimations from radiographs.
• The system functions as a valuable decision-support tool for TPLO planning, requiring surgeon verification.
• Occasional landmark detection failures necessitate continued surgeon oversight for optimal patient care.