A neural network diagnostic tool for the chronic fatigue syndrome

Artificial neural networks are particularly useful for problems which are difficult to find an algorithmic or rule-based solution. A typical example of such problems is the diagnosis of chronic fatigue syndrome (CFS). CFS patients exhibit complex patterns of multiple and varying (non-consistent) psychiatric and somatic symptoms making diagnosis very difficult. This study demonstrates that a simple feedforward neural network can be trained on a multidisciplinary questionnaire to accurately diagnose CFS. The neural network was able to clearly differentiate between CFS patients, patients suffering from major unipolar depression, lipid patients and a healthy control group. By analyzing the resultant neural connection weights an indication which symptoms differentiate CFS patients from the control groups was obtained. The largest weights were assigned to somatic symptoms like fibromyalgia, photophobia and night sweats and to fatigue related to physical activity. By excluding those questions with very small absolute weights, a shortened user-friendly self-report questionnaire (n_q =70) was obtained that can be particularly useful for primary health care