Three-dimensional visualization of human hemoglobin phenotypes with HPLC.

Hemoglobinopathies were included in the Brazilian Neonatal Screening Program on June 6, 2001. Automated high-performance liquid chromatography (HPLC) was indicated as one of the diagnostic methods. The amount of information generated by these systems is immense, and the behavior of groups cannot always be observed in individual analyses. Three-dimensional (3-D) visualization techniques can be applied to extract this information, for extracting patterns, trends or relations from the results stored in databases. We applied the 3-D visualization tool to analyze patterns in the results of hemoglobinopathy based on neonatal diagnosis by HPLC. The laboratory results of 2520 newborn analyses carried out in 2001 and 2002 were used. The "Fast", "F1", "F" and "A" peaks, which were detected by the analytical system, were chosen as attributes for mapping. To establish a behavior pattern, the results were classified into groups according to hemoglobin phenotype: normal (N = 2169), variant (N = 73) and thalassemia (N = 279). 3-D visualization was made with the FastMap DB tool; there were two distribution patterns in the normal group, due to variation in the amplitude of the values obtained by HPLC for the F1 window. It allowed separation of the samples with normal Hb from those with alpha thalassemia, based on a significant difference (p < 0.05) between the mean values of the "Fast" and "A" peaks, demonstrating the need for better evaluation of chromatograms; this method could be used to help diagnose alpha thalassemia in newborns.

A support system for content-based medical image retrieval in object oriented databases.

This work describes both the concepts used in an Object Manager for storage of medical images as one more data type associated to objects, and a support system developed to offer this kind of tool to medical application developers. The purpose of this work is to support the retrieval of images through queries based on the graphical contents of the stored images. The usual approach uses icons and textual attributes stored with the images to specify the queries. This work uses a novel modeling technique to define the "image data type," by means of which it is possible to decide, before the query itself, the key data of each image that must be extracted from the image when it is stored in the database, so the search can be accelerated when queries are issued. This approach enables building of expansible systems, where new image processing algorithms can be added easily, using its syntactic representation stored through an Image Meta-schema into the application database schema. This work shows how such a system has been implemented, and also provides a query language used to refer and execute these algorithms from inside the database management system.