Among several features used for clinical binary classification, behavioral performance, questionnaire scores, test results, and physical exam reports can be counted. Attempts to include neuroimaging findings to support clinical diagnosis are scarce due to difficulties in collecting such data, as well as problems in integration of neuroimaging findings with other features. The binary classification method proposed here aims to merge small samples from multiple sites so that a large cohort, which better describes the features of the disease can be built. We implemented a simple and robust framework for detection of fibromyalgia, using likelihood during decision level fusion. This framework supports sharing of classifier applications across clinical sites and arrives at a decision by fusing results from multiple classifiers. If there are missing opinions from some classifiers due to inability to collect their input features, such degradation in information is tolerated. We implemented this method using functional near infrared spectroscopy (fNIRS) data collected from fibromyalgia patients across three different tasks. Functional connectivity maps are derived from these tasks as features. In addition, self-reported clinical features are also used. Five classifiers are trained using k nearest neighborhood (kNN), linear discriminant analysis (LDA), and support vector machine (SVM). Fusion of classification opinions from multiple classifiers based on likelihood ratios outperformed individual classifier performances. When 2, 3, 4, and 5 different classifiers are fused, sensitivity, and specificity figures of 100% could be obtained based on the choice of the classifier set.