COMPUTATIONAL ANALYSIS BASED ON ARTIFICIAL NEURAL NETWORKS FOR AIDING IN DIAGNOSING OSTEOARTHRITIS OF THE LUMBAR SPINE

Few data exist concerning the natural history of lumbar spine disc degeneration and associated risk factors. We therefore undertook this study to examine the radiographic progression of lumbar spine disc degeneration over the course of 9 years in a population-based inception cohort of women from the Chingford Study. Seven hundred ninety-six paired lumbar spine radiographs were read by a single reader for anterior osteophytes (AO) and disc space narrowing (DSN) using the Lane atlas at each lumbar disc space (L1-5). Disc degeneration was defined using thresholds of AO and DSN grade 1+ in one or more vertebrae (L1-5) within a subject. Progression was defined as an increase in grade in an affected year-1 vertebra. Potential risk factors were assessed using odds ratios and 95% confidence intervals adjusted for age, body mass index (BMI), and other potential confounders in logistic regression models using the STATA statistical package. The mean +/- SD age at baseline was 53.8 +/- 6.0 years, and mean +/- SD BMI was 25.4 +/- 4.1 kg/m(2). Progression rates for AO and DSN were 4% per annum and 3% per annum, respectively. Progression of DSN was predicted by age, back pain, and radiographic hip and knee osteoarthritis (OA). Progression of AO was predicted by age and radiographic hip OA, with borderline significance for BMI >30 kg/m(2). No significant effects were seen for smoking, physical activity, hormone replacement therapy use, multiparity, or hand OA. This is the first population-based longitudinal study to assess progression of the individual radiographic features of AO and DSN in lumbar spine disc degeneration. We demonstrated progression rates of 3-4% per annum, with important risk factors for progression, including age, back pain, and radiographic OA at the hip and knee.

Lumbar disc degeneration is characterised radiologically by the presence of osteophytes, end-plate sclerosis and disc space narrowing. To determine the strength of the association between increasing severity of combinations of these features in a population sample of men and women. Men and women aged >or=50 years were recruited from a primary care-based community health index in Aberdeen, UK. Participants had lateral spinal radiographs performed according to a standard protocol. The intervertebral disc spaces (L1/2-L4/5) were evaluated for the presence of anterior osteophytes, end-plate sclerosis and disc space narrowing using a graded semiquantitative score (grade 0-3). Log linear modelling was used to determine the associations (pairwise) between increasing severity of these features, with the results expressed as beta coefficients and 95% confidence intervals (CIs). There were 286 men (mean age 65.3 years) and 299 women (mean age 65.2 years) with spinal radiographs, yielding a total of 2340 assessable lumbar vertebral levels. In all, 73% of vertebral levels had evidence of osteophytes, 26% of sclerosis and 37% of disc space narrowing. Increasing severity of osteophyte grade was associated with an increasing severity both of sclerosis and of disc space narrowing, whereas the severity of sclerosis was associated with the severity of narrowing. This was true at all vertebral levels. The strongest association, however, was between osteophytes and sclerosis (beta coefficient = 2.7, 95% CI 2.4 to 3.1). For sclerosis and narrowing the beta coefficient was 1.9 (95% CI 1.7 to 2.1), whereas for osteophytes and narrowing the beta coefficient was much weaker at 1.2 (95% CI 1.1 to 1.3). There was no important influence of vertebral level on any of these associations. The association between increasing severity of osteophytes and end-plate sclerosis is stronger than for other combinations of radiographic features of lumbar disc degeneration.

OBJETIVO: Avaliar as redes neurais recorrentes enquanto técnica preditiva para séries temporais em saúde. MÉTODOS: O estudo foi realizado durante uma epidemia de cólera ocorrida no Estado do Ceará, em 1993 e 1994, a partir da sobremortalidade tendo como causa básica as infecções intestinais mal definidas (CID-9). O número mensal de óbitos por essa causa, referente ao período de 1979 a 1995 no Estado do Ceará, foram obtidos do Sistema de Informação de Mortalidade (SIM) do Ministério da Saúde. Estruturou-se uma rede com dois neurônios na camada de entrada, 12 na camada oculta, um neurônio na camada de saída e um na camada de memória. Todas as funções de ativação eram a função logística. O treinamento foi realizado pelo método de backpropagation, com taxa de aprendizado de 0,01 e momentum de 0,9, com dados de janeiro de 1979 a junho de 1991. O critério para fim do treinamento foi atingir 22.000 epochs. Compararam-se os resultados com os de um modelo de regressão binomial negativa. RESULTADOS: A predição da rede neural a médio prazo foi adequada, em dezembro de 1993 e novembro e dezembro de 1994. O número de óbitos registrados foi superior ao limite do intervalo de confiança. Já o modelo regressivo detectou sobremortalidade a partir de março de 1992. CONCLUSÕES: A rede neural se mostrou capaz de predição, principalmente no início do período, como também ao detectar uma alteração concomitante e posterior à ocorrência da epidemia de cólera. No entanto, foi menos precisa do que o modelo de regressão binomial, que se mostrou mais sensível para detectar aberrações concomitantes à circulação da cólera.