Validation of a new multiple osteochondromas classification through Switching Neural Networks

Area of Science:

• Genetics and Molecular Biology
• Orthopedics
• Medical Informatics

Background:

• Multiple osteochondromas (MO), also known as hereditary multiple exostoses (HME), is an autosomal dominant condition.
• MO is characterized by the development of numerous benign cartilage-capped bone tumors (osteochondromas or exostoses).
• Existing clinical classifications for MO lack consensus, hindering patient stratification and research.

Purpose of the Study:

• To validate a user-friendly tool for classifying MO patients into three distinct clinical groups.
• To utilize a machine learning approach for objective patient characterization based on disease severity and impact.

Main Methods:

• A machine learning approach, specifically a Switching Neural Network, was employed.
• The model analyzed 150 variables across 289 MO patients.
• The classification was based on affected bone segments, skeletal deformities, and functional limitations.

Main Results:

• The proposed classification tool achieved a highly satisfactory mean accuracy.
• The study identified specific clinical features: ankle valgism, Madelung deformity, and limited hip extra-rotation, as key indicators for the three classes.
• The machine learning model provided intelligible if-then rules for classification.

Conclusions:

• The developed classification system offers an efficient method for characterizing MO.
• This tool enables the definition of homogeneous patient cohorts for future research into MO pathogenesis.
• The classification facilitates a better understanding of the clinical spectrum of this rare disease.