Best Frames Selection and Montaging in Wide-field ROP Images

Abstract

 

Purpose

 

To develop a computerized system to select the best still frames from a video acquired with wide-field fundus cameras in Retinopathy of Prematurity (ROP) infants and to build a mosaic from these images. It will allow clinicians to examine a single large, best quality image.

 

Methods

 

20 videos of retinal fundus were acquired with RetCam (Clarity Medical Systems, Pleasanton, CA, USA) in normal or ROP infants. The majority of the video frames are usually altered by defects such as artifacts, interlacing and defocus. The best frames (not marked by these defects) were identified using several image quality parameters, derived by image analysis tools specifically developed for this purpose and based on Laplacian of Gaussian filtering. The selected frames were then registered by applying rotation and translation movements to achieve the best overlapping in all the areas of intersection, until a single mosaic image was obtained. A custom blending procedure, based on various schemes of pixel intensity weighting, was then developed to provide a final image with homogeneous luminosity and contrast, devoid of the dark areas typically present in the outer regions of singles frames.

 

Results

 

5500 frames from the 20 videos were selected for the composition of the ground truth reference, where they were manually labeled as either “high quality” or “normal to poor quality”. This set has been used to validate the best-frame selection stage, which showed a PPV of 0.88 (PPV: Positive Predicted Value = fraction of true high quality over the estimated high quality). As far as registration and frame blending are concerned, simple visual inspection of the resulting mosaics confirmed the outstanding capability of the proposed system to provide higher quality images (Fig. 1,2).

 

Conclusions

 

The proposed system allowed in this dataset the successful selection and montaging of the best video frames. The resulting mosaic image provides a larger and higher quality image than single frames, which should significantly improve the clinical evaluation of ROP from fundus camera videos.

 

 

Fig. 1. Example of composed mosaic: frames blended with standard montaging procedure.

 

 

Fig. 2. Example of composed mosaic: frames blended with proposed algorithm.

 

Keywords: 549 image processing • 706 retinopathy of prematurity