Neural Networks: A Comprehensive Foundation
Neural Networks: A Comprehensive Foundation
Pattern Recognition and Neural Networks
Pattern Recognition and Neural Networks
| Hi-index | 0.00 |
Secondary lymphedema of upper limbs, a frequent complication after a breast cancer therapy, can be successfully treated only when diagnosed at an early, ideally latent, stage. Lymphoscintigraphy is a promising candidate to this purpose. A slow lymphatic dynamics of upper limbs allows, however, a routine collection at most three images reflecting it. This makes an exploitation of lymphoscintigraphy to early-stage diagnosis a complex task. Recently, a Bayesian methodology extracting diagnostic information from the available sparse data has been developed. It properly detects lymphedema occurrence but not a desirable disease staging. The present paper proposes Bayesian diagnostic processing of lymphoscintigraphic and routine clinical data. Its staging ability was tested on diagnostic data of 88 women at the age of 39-84 years (60.2+/-10.4) with a suspicion of unilateral secondary lymphedema of upper limbs caused by a breast cancer treatment. Less than 20 of them had simply detectable disease stages. Information about accumulation dynamics of the lymphatic system contained in lymphoscintigraphic images was quantified via estimation of a simplified accumulation model [P. Gebousky, M. Karny, A. Quinn, Lymphoscintigraphy of upper limbs: a Bayesian framework, in: J.M. Bernardo, M.J. Bayarri, J.O. Berger (Eds.), Bayesian Statistics, vol. 7, University Press, Oxford, 2003, pp. 543-552]. The sole use of this approach, referred as ''Bayesian quantitative lymphoscintigraphy'', was found insufficient for a finer staging of the disease to typical categories (healthy, latent, reversible, spontaneously irreversible, elephantiasis). For this reason, the results of Bayesian quantitative lymphoscintigraphy were attached to routinely available qualitative lymphoscintigraphic inspection and clinical data. These combined data were modelled by normal probabilistic mixtures. Their Bayesian estimates were used for a computerized disease staging. The resulting model predicts expert's conclusions on the presence of a lymphedema in 95% cases. A finer staging is successful in 85% cases of suspicious limbs. Model cross-validation and a closer look on patients' data indicate that the combined data are still insufficiently informative. It calls for the further improvements of the inspection methods. Even under the current inspection conditions, the proposed processing provides clinicians a reliable quantitative ''second'' opinion on the disease staging.