AUTOMATED DIAGNOSIS OF LOW BACK PAIN

Abstract

Introduction: Delay in active management reduces the prognosis for simple low back pain. The aim of this project was to develop a tool for use in GP surgeries to assist the doctor in his/her diagnosis of lower back pain and allow prompt management with confidence.

Methods: Three different systems for the automated diagnosis of low back pain were developed. With each, the patient answered a series of questions presented by the system. Three different strategies were employed, one using variable weighting, one a logic tree and one an inference engine. For the purpose of testing the systems against each other, a database was constructed containing the answers to all possible questions from each system for one hundred patients attending a low back pain clinic. The “true” diagnosis was that made by the treating clinician who saw the patients.

The original data contained a number of diagnoses:

• Spinal Stenosis (central or lateral)

• Prolapsed Intervertebral Disc

• Other Nerve Root Compression (NRC)

• Mechanical Back Pain (MBP) with NRC

• Pure MBP

For the purpose of the comparison two groups were considered – patients with radicular symptoms (groups 1 to 4) and patients with pure MBP.

Conclusions: The different approaches to development showed that a number of factors play a crucial role for the accuracy of the systems, including the number of rules used to try to cover every possibility, the interpretation of the questions by the patients and the weighting and approach taken for the different Certainty Factors. The use of any of these three approaches did not allow the development of a system accurate enough for clinical use and it seems that successful development of such a system might require a wholly different approach.