No abstract

Background Alzheimer’s disease has become a public health crisis globally due to its increasing incidence. The purpose of this study was to establish an early warning model using artificial neural network (ANN) for early diagnosis of AD and to explore early sensitive markers for AD. Methods A population based nested case-control study design was used. 89 new AD cases with good compliance who were willing to provide urine and blood specimen were selected from the cohort of 2482 community-dwelling elderly aged 60 years and over from 2013 to 2016. For each case, two controls living nearby were identified. Biomarkers for AD in urine and blood, neuropsychological functions and epidemiological parameters were included to analyze potential risk factors of AD. Compared with logistic regression, k-Nearest Neighbor (kNN) and support vector machine (SVM) model, back-propagation neural network of three-layer topology structures was applied to develop the early warning model. The performance of all models were measured by sensitivity, specificity, accuracy, positive prognostic value (PPV), negative prognostic value (NPV), the area under curve (AUC), and were validated using bootstrap resampling. Results The average age of AD group was about 5 years older than the non-AD controls ( P < 0.001). Patients with AD included a significantly larger proportion of subjects with family history of dementia, compared with non-AD group. After adjusting for age and gender, the concentrations of urinary AD7c-NTP and aluminum in blood were significantly higher in AD group than non-AD group (2.01 ± 1.06 vs 1.03 ± 0.43, 1.74 ± 0.62 vs 1.24 ± 0.41 respectively), but the concentration of Selenium in AD group (2.26 ± 0.59) was significantly lower than that in non-AD group (2.61 ± 1.07). All the models were established using 18 variables that were significantly different between AD patients and controls as independent variables. The ANN model outperformed the other classifiers. The AUC for this ANN was 0.897 and the model obtained the accuracy of 92.13%, the sensitivity of 87.28% and the specificity of 94.74% on the average. Conclusions Increased risk of AD may be associated with higher age among senior citizens in urban communities. Urinary AD7c-NTP is clinically valuable for the early diagnosis. The established ANN model obtained a high accuracy and diagnostic efficiency, which could be a low-cost practicable tool for the screening and diagnosis of AD for citizens.

No abstract

The emerging availability of trained machine learning models has put forward the novel concept of Machine Learning Model Market in which one can harness the collective intelligence of multiple well-trained models to improve the performance of the resultant model through one-shot federated learning and ensemble learning in a data-free manner.However, picking the models available in the market for ensemble learning is time-consuming, as using all the models is not always the best approach. It is thus crucial to have an effective ensemble selection strategy that can find a good subset of the base models for the ensemble. Conventional ensemble selection techniques are not applicable, as we do not have access to the local datasets of the parties in the federated learning setting. In this paper, we present a novel Data-Free Diversity-Based method called DeDES to address the ensemble selection problem for models generated by one-shot federated learning in practical applications such as model markets. Experiments showed that our method can achieve both better performance and higher efficiency over 5 datasets and 4 different model structures under the different data-partition strategies.

No abstract