Objectives: The histological examination of blood vessel and nerve density in tendon tissue is laborious and potentially imprecise. Current methods of examination typically involve manual counting of structures in a prescribed region of interest. Artificial intelligence offers new possibilities to more efficiently and precisely characterise vessels and nerves in tendon tissue. The aim of this study was to develop a machine learning method to identify and quantify blood vessels and nerves within the gluteal tendons.
Methods: Transverse histological sections of gluteal tendons from consented donors (3 males and 3 females aged 58 – 93 years) underwent immunohistochemical staining for CD34 (blood vessels) and S100 (nerves). The slides were scanned to 40x magnification using a ZEISS AxioScan.Z1 (Carl Zeiss AG, Oberkochen, Germany) and analysed using QuPath V0.4.4. An automated approach using QuPath was devised to distinguish tendon from muscle tissue and identify blood vessels. The automated method was reviewed and corrected manually (semi-automated). The cross-sectional area of tendon, muscle and blood vessels was calculated and the automated and semi-automated approaches compared.
Results: Approximately 30 adjustable parameters were required to characterise tendon and muscle tissue. There was approximately 15% error between the automated and semi-automated methods for the identification and measurement of blood vessel cross-sectional area in tendon.
Conclusion: Machine learning increases the efficiency of analysing large histological datasets. However, the development of the method requires significant expert training and testing.
Keywords: innervation; machine learning; tendon; vascularity
Ethical statement: Ethical approval was granted by the Australian National University Human Ethics Committee.
Funding statement: This study was supported by the Act Health Vacation Study Program