Abstract The mining industry can be one of the most impacting human activities. In the southern region of Santa Catarina (Brazil), open pit coal mining has left an extensive environmental impact. Since there was no topsoil in the abandoned open pit sites, it is necessary to provide a substrate for vegetation growth. However, the selection of the best substrate between multiple options is difficult. Thus, a fuzzy logic-based model is proposed. The proposed model was compared to reference models and to experts’ knowledge. Statistical analysis and validation were carried out with a correlation coefficient, a Kappa coefficient, along with the Accuracy, Precision, Sensibility Specificity, F-Score and Mathews correlation coefficients. The data set used to assess the proposed model presented a wide range of data, but for values such as aluminum saturation, higher values were common. The fuzzy logic-based expert system presented better results when assessing the behavior of the defuzzified output values with the crisp input values. The fuzzy model also followed the trend of the reference models (with R2 between 0.3639 and 0.5250). The comparison to the experts’ opinion demonstrated that agreement comes easily with extreme values (such as not suitable and suitable). However, using a Winner-Takes-All approach, the proposed fuzzy model had high scores for suitable soils for land reclamation’s soil construction. The proposed model can be used to define the best substrate for land reclamation. Some improvements, such as different parameters and increases in the number of interviews rounds, should be also tested.

OBJETIVOS: Conhecer as vantagens da utilização das redes neurais artificiais no reconhecimento de padrões em radiografias de coluna lombar na incidência perfil para auxiliar no diagnóstico da osteoartrite primária. MÉTODO: Estudo transversal, descritivo, analítico, de abordagem quantitativa e com ênfase diagnóstica. O conjunto de treinamento foi composto por imagens coletadas no período de janeiro a julho de 2009 de pacientes submetidos a radiografias digitais de coluna lombar na incidência em perfil provenientes de um serviço de radiologia localizado no município de Criciúma (SC). Das 260 imagens coletadas, foram excluídas: as radiografias distorcidas, as patologias que alteram a arquitetura da coluna lombar e os padrões de difícil caracterização, resultando em um total de 206 imagens. O banco de imagens (n = 206) foi subdividido, resultando em 68 radiografias para a etapa de treinamento, 68 para testes e 70 para validação. Foi utilizada uma rede neural híbrida baseada em mapas auto-organizáveis de Kohonen e redes Multilayer Perceptron. RESULTADOS: Após 90 ciclos, foi realizada a validação com o melhor teste, alcançando acurácia de 62,85%, sensibilidade de 65,71% e especificidade de 60%. CONCLUSÃO: Apesar da demonstração de uma eficácia mediana, por se tratar de estudo de caráter inovador, seus valores mostram um futuro promissor da técnica utilizada, com sugestão para trabalhos futuros com abrangência na metodologia de processamento das imagens e ciclos com uma quantidade maior de radiografias.

OBJECTIVE: To ascertain the advantages of applying artificial neural networks to recognize patterns on lumbar column radiographs in order to aid in the process of diagnosing primary osteoarthritis. METHODS: This was a cross-sectional descriptive analytical study with a quantitative approach and an emphasis on diagnosis. The training set was composed of images collected between January and July 2009 from patients who had undergone lateral-view digital radiographs of the lumbar column, which were provided by a radiology clinic located in the municipality of Criciúma (SC). Out of the total of 260 images gathered, those with distortions, those presenting pathological conditions that altered the architecture of the lumbar column and those with patterns that were difficult to characterize were discarded, thus resulting in 206 images. The image data base (n = 206) was then subdivided, resulting in 68 radiographs for the training stage, 68 images for tests and 70 for validation. A hybrid neural network based on Kohonen self-organizing maps and on Multilayer Perceptron networks was used. RESULTS: After 90 cycles, the validation was carried out on the best results, thereby reaching accuracy of 62.85%, sensitivity of 65.71% and specificity of 60%. CONCLUSIONS: Even though the effectiveness shown was moderate, this study is of innovative nature. Hence, the values show that the technique used has a promising future, thus pointing towards further studies covering the image and cycle processing methodology with a larger quantity of radiographs.