Comparison of Gross Pathology, Histopathology, and Mycobacterial Culture for the Diagnosis of Tuberculosis in Elk (Cervus elaphus)

Using the isolation of Mycobacterium bovis as the reference standard, this study evaluated the sensitivity, specificity and kappa statistic of gross pathology (abattoir postmortem inspection), histopathology, and parallel or series combinations of the two for the diagnosis of tuberculosis in 430 elk and red deer. Two histopathology interpretations were evaluated: histopathology I, where the presence of lesions compatible with tuberculosis was considered positive, and histopathology II, where lesions compatible with tuberculosis or a select group of additional possible diagnoses were considered positive. In the 73 animals from which M. bovis was isolated, gross lesions of tuberculosis were most often in the lung (48), the retropharyngeal lymph nodes (36), the mesenteric lymph nodes (35), and the mediastinal lymph nodes (16). Other mycobacterial isolates included: 11 M. paratuberculosis, 11 M. avium, and 28 rapidly growing species or M. terrae complex. The sensitivity estimates of gross pathology and histopathology I were 93% (95% confidence limits [CL] 84,97%) and 88% [CL 77,94%], respectively, and the specificity of both was 89% [CL 85,92%]). The sensitivity and specificity of histopathology II were 89% (CL 79,95%) and 77% (CL 72,81%), respectively. The highest sensitivity estimates (93- 95% [CL 84,98%]) were obtained by interpreting gross pathology and histopathology in parallel (where an animal had to be positive on at least one of the two, to be classified as combination positive). The highest specificity estimates (94-95% [CL 91-97%]) were generated when the two tests were interpreted in series (an animal had to be positive on both tests to be classified as combination positive). The presence of gross or microscopic lesions showed moderate to good agreement with the isolation of M. bovis (Kappa = 65-69%). The results show that post-mortem inspection, histopathology and culture do not necessarily recognize the same infected animals and that the spectra of animals identified by the tests overlaps.

An Extended Kernel for Generalized Multiple-Instance Learning

The multiple-instance learning (MIL) model has been successful in areas such as drug discovery and content-based image-retrieval. Recently, this model was generalized and a corresponding kernel was introduced to learn generalized MIL concepts with a support vector machine. While this kernel enjoyed empirical success, it has limitations in its representation. We extend this kernel by enriching its representation and empirically evaluate our new kernel on data from content-based image retrieval, biological sequence analysis, and drug discovery. We found that our new kernel generalized noticeably better than the old one in content-based image retrieval and biological sequence analysis and was slightly better or even with the old kernel in the other applications, showing that an SVM using this kernel does not overfit despite its richer representation.