A Case-Based Reasoning Approach to Mental State Examination Using a Similarity Measure Based on Orthogonal Vector Projection

Mental state examination (MSE) involves assessing the overall severity of illness and also differentiating likely diagnoses. When it is performed on patients serially during the period of their illness, the consecutive estimates can serve as an important way to track their recovery. However, the traditional approach to mental state examination that uses clinician's subjective judgement and results in subjective estimates can be unreliable and prone to inconsistencies. Using the approach introduced in this paper, a case is represented as a vector of thirty five different clinical features, which are rated using a numerical scale according to the severity of each clinical feature. The vector length is used as a measure of the overall severity of the illness. The case base consists of one standard case for each of 6 diagnostic categories. Each standard case represents a typical case for its diagnostic category, with each clinical feature rated according to the maximum level of severity that can be expected for that category. Evaluation of a given clinical case, with clinical features as rated by a clinician with regard to the likely diagnoses involves measuring the similarity of the resulting case vector with the standard vectors in the case base. Whilst cosine similarity and Euclidean distance are alternative measures of similarity, a more clinically intuitive and accurate measure based on orthogonal vector projection is proposed. The orthogonal vector projection approach to case based assessment was evaluated using thirty different test cases representing six different common diagnostic categories. For each of the test cases similarity measures obtained using orthogonal vector projection were compared with measures obtained using cosine similarity and Euclidean distance. The results indicated that the orthogonal vector projection approach was able to differentiate both the diagnosis and severity of illness more accurately than the other two similarity measures. The proposed approach has the potential to be used as a standardised clinical tool for both establishing the diagnosis and severity of illness, and also measuring the recovery from illness. In particular, the estimates of recovery obtained from this approach can serve as an important index in healthcare economics.