Assessing tuberculosis in the skeleton with the use of decision tree analysis

Area of Science:

• Paleopathology
• Medical Informatics
• Forensic Anthropology

Background:

• Diagnosing skeletal infectious diseases is challenging, often relying on subjective expert opinion.
• Statistical methods are underutilized in skeletal disease diagnosis.
• Tuberculosis (TB) can manifest with distinct skeletal lesions, but recognition in skeletal remains requires systematic analysis.

Purpose of the Study:

• To apply data mining and machine learning, specifically decision tree analysis, to identify patterns of skeletal involvement in tuberculosis (TB).
• To develop a model that assists in recognizing TB in skeletal remains.
• To explore the utility of computational methods in diagnosing skeletal diseases.

Main Methods:

• A dataset of 387 modern South African individuals (207 with TB, 180 controls) was analyzed.
• Presence or absence of 21 skeletal lesions associated with TB was scored.
• A pruned decision tree classification analysis was performed to identify significant patterns.

Main Results:

• A decision tree model with moderate classification accuracy was developed using four key variables.
• Vertebral changes were the most significant predictor, followed by rib, acetabular, and cranial lesions.
• Machine learning successfully differentiated skeletal patterns between TB-affected and control groups.

Conclusions:

• Machine learning, via decision tree analysis, can identify patterns indicative of skeletal tuberculosis.
• Vertebral, rib, acetabular, and cranial lesions are important indicators in TB diagnosis from skeletal remains.
• Further research is warranted to explore machine learning's broader application in diagnosing skeletal diseases.