A statistical model for ultrasound diagnosis of soft-tissue tumours in the hand and forearm.

Abstract

The ultrasound characteristics of 5 common tumours of the hand and forearm were used to build a statistical model that could assess the weight of each trait or combination of traits in the correct diagnosing of the tumours. The model was used for calculating the first and second diagnostic alternatives. The basic data of the model were the ultrasound findings in previously presented material on the 5 common benign soft-tissue tumours. The material consisted of 96 tumours: 18 villonodular synovites, 26 haemangiomas, 14 lipomas, 27 nerve tumours, and 11 glomus tumours. To build the statistical model, the ultrasound characteristics that were significant at 5% level were calculated. With stepwise logistic regression, 3 of these were selected as explaining variables. The degree of influence of the explaining variable on the response variable was calculated by way of odds quotients. The material was then analysed by means of the statistical model. The correct diagnosis was calculated as first and second alternatives for each tumour and for the whole material. The diagnostic accuracy for the whole material was 51% in the first choice and 77% in first plus second choices. Ultrasound should be the first imaging modality for soft-tissue tumours of the hand. However, MR should also be performed when diagnosis continues to be obscure and when malignancy is suspected.