Nonpooling Convolutional Neural Network Forecasting for Seasonal Time Series With Trends.

This article focuses on a problem important to automatic machine learning: the automatic processing of a nonpreprocessed time series. The convolutional neural network (CNN) is one of the most popular neural network (NN) algorithms for pattern recognition. Seasonal time series with trends are the most common data sets used in forecasting. Both the convolutional layer and the pooling layer of a CNN can be used to extract important features and patterns that reflect the seasonality, trends, and time lag correlation coefficients in the data. The ability to identify such features and patterns makes CNN a good candidate algorithm for analyzing seasonal time-series data with trends. This article reports our experimental findings using a fully connected NN (FNN), a nonpooling CNN (NPCNN), and a CNN to study both simulated and real time-series data with seasonality and trends. We found that convolutional layers tend to improve the performance, while pooling layers tend to introduce too many negative effects. Therefore, we recommend using an NPCNN when processing seasonal time-series data with trends. Moreover, we suggest using the Adam optimizer and selecting either a rectified linear unit (ReLU) function or a linear activation function. Using an NN to analyze seasonal time series with trends has become popular in the NN community. This article provides an approach for building a network that fits time-series data with seasonality and trends automatically.

Laboratory validation studies in Ki-67 digital image analysis of breast carcinoma: a pathway to routine quality assurance.

Ki-67 proliferative index (PI) has prognostic and predictive value in invasive breast carcinoma (IBC), but clinical uptake has been hampered by suboptimal accuracy, reproducibility and standardisation. Published guidelines have addressed pre-analytical and analytical factors to improve Ki-67 PI utility; however, practicalities of ongoing monitoring of Ki-67 PI quality in IBC reporting have not been established. We aimed to evaluate the internal and external quality of our established digital Ki-67 PI IBC reporting practice at a tertiary institution. In the 5 years since initial validation work, we've completed a series of internal and external quality assurance (QA) projects: (1) an interobserver agreement study, (2) a two site interlaboratory agreement study, (3) determination of the error of our Ki-67 results, (4) an audit of the year-to-year Ki-67 values, (5) an audit of Ki-67 in neoadjuvant chemotherapy (NAC) treated cases, and (6) comparison of our Ki-67 datasets with similar published datasets. There was excellent concordance (intra-class correlation = 0.98) and good agreement [kappa (κ) = 0.76-0.96] between pathologists, excellent concordance [Pearson correlation (R) = 0.94] and very good agreement (κ = 0.80) between laboratories and excellent concordance (R = 0.92-0.95) and good agreement (κ = 0.67-1.0) over time for our Ki-67 results. No significant difference was observed in Ki-67 data from year-to-year. Expected associations with clinico-pathological prognosticators, pathological complete response following NAC and mitotic index were evident. The median Ki-67 values from the overall and NAC treated datasets were within the range reported in other studies, and our data could be separated into similarly proportioned 'high' and 'low' Ki-67 PI groups when dichotomised as per protocols in other studies. Collectively, our work provides evidence of adequate internal and external quality control for our digital Ki-67 PI IBC reporting protocols. Given the paucity of formal Ki-67 QA programs, our approach could be emulated, and results compared between laboratories as a framework for internal and external Ki-67 QA.

Sonification of optical coherence tomography data and images.

Sonification is the process of representing data as non-speech audio signals. In this manuscript, we describe the auditory presentation of OCT data and images. OCT acquisition rates frequently exceed our ability to visually analyze image-based data, and multi-sensory input may therefore facilitate rapid interpretation. This conversion will be especially valuable in time-sensitive surgical or diagnostic procedures. In these scenarios, auditory feedback can complement visual data without requiring the surgeon to constantly monitor the screen, or provide additional feedback in non-imaging procedures such as guided needle biopsies which use only axial-scan data. In this paper we present techniques to translate OCT data and images into sound based on the spatial and spatial frequency properties of the OCT data. Results obtained from parameter-mapped sonification of human adipose and tumor tissues are presented, indicating that audio feedback of OCT data may be useful for the interpretation of OCT images.

Epidemic simulation for syndromic surveillance.

This article reports on a project to develop a simulation-based test bed for the BioDefend Syndromic Surveillance System. BioDefend is a system that data mines syndrome reports from emergency rooms and so forth to produce early alerts of epidemic onset. An existing large-scale epidemic simulation will be adapted to provide synthetic reports of syndromes associated with extremely rare events such as pandemics and bioterrorism. The Spatiotemporal Epidemiological Modeler will be used as the basis of the test bed. Results from the much simpler Spatiotemporal Epidemiological Modeler simulation will be validated by comparison against results from the more complex Epidemiological Simulation System. These synthesized reports will be used to test BioDefend's ability to detect epidemic outbreaks and to evaluate its data-mining algorithm. The development of an optimal algorithm for processing syndrome reports to provide reliable epidemic early warnings is a difficult research problem that the test bed should help address.

Test for the consistency of noninferiority from multiple clinical trials.

A testing procedure is proposed to assess the consistency of noninferiority from a collection of trials based on simultaneous t lower confidence bounds or Scheffé's lower confidence bounds. Methods for simultaneous inferences on pairwise or many-to-one comparisons among multiple noninferiority trials are also discussed. To avoid bias due to subjective trial exclusion a tuning parameter k is embedded into the testing procedure to provide flexibility to quantify the "consistency of noninferiority" when the total number of trials is large. The size and power of the proposed test are discussed. The method is illustrated using simulations and real data analysis.

Is the curve relating temperature to aggression linear or curvilinear? Assaults and temperature in minneapolis reexamined.

Using archival data from Minneapolis recorded in 3-hr time intervals, E. G. Cohn and J. Rotton concluded that there is an inverted U-shaped relationship between temperature and assault, with the maximum assault rate occurring at 74.9 degrees F. They depicted this relationship by plotting temperature against assault. This plot, however, fails to take into account time of day. Time of day was strongly related to both temperature and assault, but in opposite directions. Between 9:00 p.m. and 2:59 a.m. of the next day, when most assaults occurred, there was a positive linear relationship between temperature and assault. The Minneapolis data actually provide stronger support of a positive linear (or monotonic) relationship between temperature and assault than of an inverted U-shaped relationship.

Is the curve relating temperature to aggression linear or curvilinear? A response to Bell (2005) and to Cohn and Rotton (2005).

P. Bell recommended examining the relationship between temperature and assaults during the hottest times of day and during the hottest months of the year. The authors' analyses of these data show a linear rather than inverted U-shaped relationship between temperature and assault during the hottest times of day and in the hottest months of the year. E. Cohn and J. Rotton recommended analyzing the 6 hr with the highest assaults versus the 6 hr with the lowest assaults. During high assault periods, there is a strong positive linear relationship between temperature and assault. During low assault periods, there is no relationship between temperature and assaults. Assaults and other violent crimes might decrease when temperatures are very hot, but the Minneapolis data set does not allow for testing of this hypothesis because Minneapolis is too cold.