Article Text
Abstract
Aims To develop a rule-based decision support system (RDSS) to assist in the diagnosis of glomerular diseases on the basis of clinical, histological and immunohistological findings.
Methods The computer software for the RDSS was written by one of the authors. The prototypical clinical and histomorphological features, including immunofluorescence findings of various glomerular diseases, were selected by three pathologists with experience in renal pathology. For every case included, all the relevant features were fed into the RDSS in the form of one or more vectors. Diagnosis was achieved by the RDSS on the basis of shortest Euclidean distance between the vector of features of a particular ‘test’ case and vectors of the prototypical features. The diagnoses rendered by the RDSS were compared with the final diagnosis signed out by the renal pathologists; the percentage of correct diagnoses was calculated for the RDSS.
Results A total of 612 cases of glomerular diseases were included in the analysis. The RDSS developed in this study gave the correct diagnosis in 83.2% of the cases included. Of all the cases, membranous glomerulonephritis and lupus nephritis were most frequently diagnosed accurately. However, immunotactoid glomerulonephritis was misdiagnosed due to the lack of ultrastructural findings in the study.
Conclusion The decision support system developed in the present study holds promise, given the high accuracy in diagnosis of glomerular diseases on the basis of clinical, histological and immunofluorescence features. Such a system may prove useful not only for diagnosis but also for postgraduate teaching and self-assessment. Results need to be confirmed in further larger studies.
- Glomerular diseases
- decision support system
- Euclidean distance