Automatic Hyphae Detection Based On Images Recognition In Fungal Keratitis Clinical Trial

Objectives:The purposes of this study were to evaluate the sensitivity and specificity of two methods in the identification of fungal keratitis,one method was automatic hyphae detection of images recognition and the other was the method of smear.To analyze the consistency of clinical symptoms and the density of hyphae which was quantified by the technology of automatic hyphae detection of images recognition.To analyze the correlation between hyphae density and prognosis.Methods:82 patients in our hospital from February 2014 to December 2019 was included in this study.All patients underwent the routine eye examination,and corneal smears(10%KOH wet method),microbial culture(blood agar and Sabouraud agar),corneal confocal microscopy,confocal microscope image recognition system examination before drug treatment.Computer image recognition system detects confocal microscope images and recognizeshypha pictures based on image recognition.56 cases were diagnosed as fungal keratitis.They all received the daily examination such as slitlamp examination and periodic review of confocal microscopy to observe the changes of corneal ulcer healing and hypha density.At first,we analyse if there were statistical differences between the four detections in the positive rate of detection of hyphae.The detection of confocal microscopy and the analysis of the automatic detection technology methods were analysed by AUC of ROC curve.We analysed the difference of disease severity,the hyphal density and prognostic evaluation between the different patients according the fungal species.At last,we compared the difference of disease severity,methods of treatment and prognostic evaluation between the different patients according the hyphal density grading.Results:According the four kinds of pathogenic examination method,38 cases were detected by corneal smear examination,the positive rate was 67.86%.42cases were detected fungus by fungal culture,the positive rate was 75.0%.53 cases were detected fungal filaments by corneal confocal microscope,the positive rate was 94.64%.50 cases were detected by the automatic image recognition technology,the positive rate of 89.29%.In the four methods mentioned before,there was no difference between corneal smears examination and fungal culture(P>0.05).But the difference between the corneal smears examination and others were statistically(P<0.05).There was statistical difference between fungal culture and corneal confocal microscopy detection and corneal confocal microscopy images automatic detection system(P<0.05).The latter is superior to the former.But there was no difference between the corneal confocal microscopy detection and corneal confocal microscopy images automatic detection system(P>0.05).The sensitivity of corneal confocal microscopy images automatic detection system is 89.29%.The specificity of corneal confocal microscopy images automatic detection system is 95.65%.And the area under the ROC curve was 0.946,while the confocal microscopy detection method of AUC was 0.951,both were above 0.900.It shows that two methods are both in high accuracy.The hyphae density can be quantitied by corneal confocal microscopy images automatic detection system according to the images detection.The score of correlation coefficient in patients between the corneal ulcer hyphae density and the clinical symptom severity was 0.8700(P<0.01).It indicates that there is a correlationship between the two methods.Different levels of hyphae density has different prognosis,low density with good prognosis while high density with poor prognosis(P<0.05).Conclusions:The technology of automatic hyphae detection of image recognition has high sensitivity and specificity in identifying fungal keratitis.The advantage of this technology is accurate,stable and without relying on the people's experience comparing with conventional artificial identification of the corneal images of confocal microscope.It is most useful to the doctors who are not familiar with fungal keratitis.The technology of automatic hyphae detection of image recognition can quantify the hyphae density and grade it.It was a kind of noninvasive examination.It can provide an evaluation criterion to fungal keratitis timely,accurately,objectively and quantitatively.It has the great positive significance for guiding the treatment of fungal keratitis.