Guimarães Sa Correia, Pedro (2017) Image Processing in Ophthalmology Towards Improved Diagnostic Markers. [Ph.D. thesis]
Questa è la versione più aggiornata di questo documento.
Full text disponibile come:
Statistiche DownloadSimple MetadataFull MetadataEndNote Format
The eye is a window to the outside world. It takes in light from the environment, converts it to electrical signals, and forwards it to the brain. In a similar manner it has been regarded as a window to our health. By imaging the eye we can visualize blood vessels and nerves noninvasively. Thus, ophthalmic imaging may allow us to diagnose and monitor both ocular and systemic diseases.
The retina has been the main focus of this line of research. Nevertheless, within the eye other structures of interest exist. Blood vessels of the bulbar conjunctiva and nerve fibers from the corneal subbasal plexus can be easily imaged. However, studies on these structures are still few in comparison to the retina. In this work, we have developed novel software tools to explore and aid the exploration of these alternative research fields.
Conjunctival Blood Vessels Segmentation
The bulbar conjunctiva is a thin mucous membrane that runs over the anterior sclera. Due to its location and transparency, this highly vascularized tissue can be easily imaged. Vascular features such as width, tortuosity or branching angle, computed from conjunctiva vasculature, may lead to important predictors of disease. Furthermore, blood vessel detection in the conjunctiva can serve as a base towards blood flow velocity computation.
We propose a novel algorithm capable of segmenting blood vessels from conjunctival images. Vessel segmentation is achieved using the phase-congruency, a dimensionless quantity invariant to contrast and scale.
In our pilot experiments, a total of 40 slit-lamp microscopy images from the conjunctiva were used test the proposed algorithm. Most blood vessels from these images appear to be correctly traced. The low running time of the proposed algorithm is appropriate for the clinical setting.
Blood Flow Velocity from Video Recordings
The use of video cameras attached to microscopes has become common. Furthermore, in vivo measurements of blood flow velocity may provide novel and/or improved disease markers. Therefore, a software tool designed to measure the blood velocity from video recordings is of interest. In this work, such software is proposed.
The software tackles three main issues in velocity measurement from videos, the registration, segmentation, and finally the measuring itself. To quantitatively validate it, a computer simulation of blood flow was created. The software was also tested in videos of an animal model (chick chorioallantoic membrane). These were captured in vivo with different resolutions, frame rates, and even cameras. The proposed algorithm performed well in all our tests. The obtained results show the robustness achieved.
Corneal Nerves Segmentation
The automatic tracing of corneal nerves is an important step for the quantitative analysis of corneal nerves in daily clinical practice. The rationale of the proposed method is to minimize required computing time while still providing accurate results. Our method consists of two sequential steps – a thresholding step followed by a supervised classification. For the classification we use a support vector machines approach. The proposed fast technique allows features, such as corneal nerve density and tortuosity, to be computed in a few seconds.
To validate the obtained tracings, we evaluated the tracing accuracy and reliability of extracted clinical parameters (corneal nerves density and tortuosity). The proposed algorithm proved capable to correctly trace 0.89 ± 0.07 of the corneal nerves. The obtained performance level was comparable to a second human grader. Furthermore, the proposed approach compares favorably to other methods. For both evaluated clinical parameters the proposed approach performed well. An execution time of 0.61 ± 0.07 seconds per image was achieved. The proposed algorithm was applied successfully to mosaic images, with run times of the order of tens of seconds.
The achieved quality and processing time of the proposed method appear adequate for the application of this technique to clinical practice. The application of nerve tracing to mosaics covering a large area can be a key component in clinical studies aimed at investigating neuropathy influence in various ocular or systemic diseases.
Corneal Nerves Density Estimation: Application to Mosaic Images
The corneal nerve density is already an established disease marker. However, most studies have only relied on single confocal images to establish correlations to diseases. The problem is that the typical confocal image only covers a small area of the central cornea. In this work we explore the corneal nerve density as computed from wide-field mosaic images.
The corneal nerves density was automatically computed and compared to a manual approach. A Pearson correlation coefficient of 0.94 was achieved. Furthermore, the proposed automatic approach was around 100 times faster (10 to 50 seconds per mosaic) than manual one.
Nerve density analysis from mosaic images was compared to the combination of multiple single images. Both underestimation and overestimation bias were verified. Furthermore, the combination of multiple single images carries a large variability depending on the area of the cornea sampled. The obtained results, demonstrate the potential of fully-automatic corneal nerve density analysis from mosaic images.
Corneal Nerves Tortuosity
The correlation between corneal nerve tortuosity and pathology has been shown multiple times. However, because there isn’t any de facto definition of tortuosity, reproducibility is poor. Indeed, many studies still rely on the manual observation and judgment of tortuosity.
Recently, two distinct forms of corneal nerve tortuosity have been identified, describing either short-range or long-range directional changes. In this study we were able to develop automatic corneal nerve tortuosity measurements that correctly and independently represent these two tortuosity definitions.
We show that a combination of mathematical tortuosity measurements improves on single metric results. Spearman rank correlation coefficients of 0.99 and 0.96 with respect to the short-range and long-range tortuosity ground-truths were obtained. These results show that the proposed tortuosity measurements highly correlate to clinical perception.
Versioni disponibili di questo documento
Solo per lo Staff dell Archivio: Modifica questo record