Super-Thresholding: Supervised Thresholding of Protein Crystal Images

Abstract

In general, a single thresholding technique is developed or enhanced to separate foreground objects from background for a domain of images. This idea may not generate satisfactory results for all images in a dataset, since different images may require different types of thresholding methods for proper binarization or segmentation. To overcome this limitation, in this study, we propose a novel approach called “super-thresholding” that utilizes a supervised classifier to decide an appropriate thresholding method for a specific image. This method provides a generic framework that allows selection of the best thresholding method among different thresholding techniques that are beneficial for the problem domain. A classifier model is built using features extracted priori from the original image only or posteriori by analyzing the outputs of thresholding methods and the original image. This model is applied to identify the thresholding method for new images of the domain. We performed our method on protein crystallization images, and then we compared our results with 6 thresholding techniques. Numerical results are provided using 4 different correctness measurements. Super-thresholding outperforms the best single thresholding method around 10%, and it gives the best performance for protein crystallization dataset in our experiments.

Applications of Artificial Intelligence and Machine Learning Algorithms to Crystallization

“…Thus, a superthresholding method has been proposed to fit the best thresholding method to individual images through a supervised classifier. Feature extraction can be performed both a priori (i.e., from the original image) or a posteriori (i.e., after applying a series of thresholding methods and mapping the processed images to the original ones) …”

“…Automated image segmentation: (a) original snapshot, (b) foreground object, (c) outer pixels, (d) inner pixels, (e) after-threshold image, and (f) inner and outer edges of the foreground image . Reproduced with permission from ref . Copyright 2017 Institute of Electrical and Electronics Engineers.…”

“…This algorithm performs well in many cases, even though the nai ̈ve assumption is rarely true. Nai ̈ve Bayes was applied, for example, as a classifier filtering the most important features for the prediction of protein crystallization propensity, 44 protein crystallization, 45 and the classification of protein crystallization images 46 and quality control of drug substances using Raman spectra. 47 It tends to handle high dimensional data well, is easily parallelizable, and can perform better than more complicated models in cases where the data set is small.…”

“…Feature extraction can be performed both a priori (i.e., from the original image) or a posteriori (i.e., after applying a series of thresholding methods and mapping the processed images to the original ones). 46 After segmentation, various ML methods can be employed to classify the crystallization outcome in an automated fashion. An interesting way to minimize the errors in the classification step is to use a decision tree classifier called "else-tree" where the else branch is followed each time the classifier cannot associate the crystallization outcome with a specific class.…”

Efficient document image binarization using heterogeneous computing and parameter tuning

“…Several of these features were inspired by the work of Dinç et al [8], in which the authors select a binarization algorithm to binarize protein crystal images based on the image's features. Examples for those features are the mean intensity of the foreground region and the standard deviation of the background region.…”