Discrimination and quantification of scar tissue by Mueller matrix imaging with machine learning

Scarring is one of the biggest areas of unmet need in the long-term success of glaucoma filtration surgery.Quantitative evaluation of the scar tissue and the post-operative structure with micron scale resolution facilitates development of anti-fibrosis techniques.However,the distinguishment of conjunctiva,sclera and the scar tissue in the surgical area still relies on pathologists'experience.Since polarized light imaging is sensitive to anisotropic properties of the media,it is ideal for discrimination of scar in the subconjunctival and episcleral area by characterizing small differences between proportion,organization and the orientation of the fibers.In this paper,we defined the conjunctiva,sclera,and the scar tissue as three target tissues after glaucoma filtration surgery and obtained their polarization characteristics from the tissue sections by a Mueller matrix microscope.Discrimination score based on parameters derived from Mueller matrix and machine learning was calculated and tested as a diagnostic index.As a result,the discrimination score of three target tissues showed significant difference between each other(p<0.001).The visualization of the discrimination results showed significant contrast between target tissues.This study proved that Mueller matrix imaging is effective in ocular scar discrimination and paves the way for its application on other forms of ocular fibrosis as a substitute or supplementary for clinical practice.

Ground subsidence mechanism of a filling mine with a steeply inclined ore body

Long-term field monitoring finds that serious surface subsidence can still occur even if the high strength cemented fill method is adopted.Combining the results of numerical simulations with global position system(GPS)monitoring,we took a typical filling mining mine with a steeply inclined ore body as an example,and explored its ground subsidence mechanism.The results show that the ground subsidence caused by the mining of steep ore body is characterized by two settlement centers and a significantly uneven spatial distribution,which is visibly different from ground subsidence characteristic of the coal mine.The subsidence on the hanging wall is much larger than that on the footwall,and the settlement center tends to move to the hanging wall with the increase of mining depth.The backfill improves the strength and surrounding rock bearing capacity,which leads to a lag of about 3 years of the subsidence.However,under the actions of continuous and repeated mining disturbances,the supporting effect of the backfill can only reduce the amplitude of the deformation,but it cannot prevent the occurrence of settlement.

Cloud Based Monitoring and Diagnosis of Gas Turbine Generator Based on Unsupervised Learning

The large number of gas turbines in large power companies is difficult to manage.A large amount of the data from the generating units is not mined and utilized for fault analysis.This study focuses on F-class(9F.05)gas turbine generators and uses unsupervised learning and cloud computing technologies to analyse the faults for the gas turbines.Remote monitoring of the operational status are conducted.The study proposes a cloud computing service architecture for large gas turbine objects,which uses unsupervised learning models to monitor the operational state of the gas turbine.Faults such as chamber seal failure,load abnormality and temperature anomalies in the gas turbine system can be identified by using the method,which has an accuracy of 60%–80%.