Initial-impression diagnosis using low-back pain patient pain drawings

Area of Science:

• Orthopedics
• Medical Imaging Analysis
• Artificial Intelligence in Medicine

Background:

• Low-back pain is a prevalent condition with diverse etiologies.
• Accurate diagnosis of lumbar spine disorders relies on various clinical assessments.
• Patient pain drawings are a subjective tool used in pain localization.

Purpose of the Study:

• To evaluate the diagnostic reliability of patient pain drawings for lumbar spine disorders.
• To compare the classification accuracy of physicians versus computerized methods (artificial neural networks).
• To identify distinct pain mark patterns associated with specific lumbar spine disorders.

Main Methods:

• Blind selection of patient pain drawings from five predefined lumbar spine disorder categories.
• Classification of drawings by experienced low-back physicians.
• Classification using discriminant analysis and artificial neural network (ANN) configurations.
• Comparative analysis of physician accuracy and ANN classification agreement.

Main Results:

• Physicians achieved an average accuracy of 51%, with noted individual biases towards certain disorders.
• Computerized methods demonstrated comparable accuracy (48%) and superior inter-rater reliability.
• Significant associations were observed between expert-predicted pain patterns and ANN-generated patterns.
• Variances in pattern associations provide insights for diagnostic prediction.

Conclusions:

• Patient pain drawings, when analyzed computationally, offer a reliable adjunct for diagnosing low-back disorders.
• Artificial neural networks provide consistent classification of pain patterns, complementing physician expertise.
• Understanding pattern variations aids clinicians in improving diagnostic accuracy from pain drawings.