APÉNDICE I PLANOS ALTERNATIVA 1 I PLANOS ALTERNATIVA
INCONVENIENTES DEL ACERO
This project rests on the hypothesis that the vasculature in tumors is substantially different from normal normal tissue, and that this difference can be imaged and quan- tified using acoustic angiography contrast-enhanced ultrasound imaging. Support for this hypothesis stems from previous work by UNC neurosurgeon Elizabeth Bullitt and work done in Dr. Paul Dayton’s laboratory. Dr. Bullitt’s work in MRA used the same implementation of the distance metric and sum of angles metric that has been presented in this chapter, and her work showed that glioblastoma multiforme (GBM) tumors had more tortuous vasculature than healthy the brain vasculature of healthy in- dividuals [13,84]. The work by Bullitt and Aylward to produce efficient segmentation of tube-like objects, define 3D tortuosity metrics, and characterize tumor and normal tor- tuosity in the brain were invaluable for informing the design and implementation of this project. Additionally, previous work in the Dayton laboratory and collaboration with Dr. Stuart Foster’s laboratory at the Sunnybrook Health Science Centre (Toronto, ON, Canada) led to the development of the confocal, dual-element transducers that were used in this work, and they also showed that superharmonic contrast imaging (acoustic angiography) could be used to measure higher average tortuosity for large, subcuta- neous fibrosarcoma tumors in rats using the same metrics [143, 144]. These preceding studies have enabled the work in this thesis, which is the first demonstration of serial vascular imaging in a spontaneous tumor model (the C3(1)/Tag) and delves deeper into understanding and characterizing vascular morphology in small tumors.
Two to three millimeters is generally considered to be the limiting size for a tumor to remain avascular [32], so it is a encouraging finding that the 2-3 mm tumors imaged in this study did indeed possess significantly higher than normal tortuosity using both tortuosity metrics. Angiogenesis begins at the capillary scale, as sprouting is typically initiated in capillaries and small, post-capillary venules [174]. Acoustic angiography,
like most non-invasive, in vivo imaging techniques cannot resolve vasculature at this scale (< 30µm) [175, 176]. Therefore, the result that 2-3 mm tumors have elevated tortuosity compared to controls in acoustic angiography images is significant because it indicates that substantial vascular remodeling in vessels larger than 100 µm has occurred by the time tumors are 2-3 mm in diameter. Angiogenesis at the capillary scale can be inferred due to the enhanced contrast signal originating in these sub-resolution vessels (and was quantified as vascular density), but the structure of individual vessels is unknown below the resolution limit. Therefore, the results presented in this chapter reinforce the fact that cancer angiogenesis produces vascular remodeling and abnormal morphology across a wide range of scales, and is not only restricted to the capillary scale.
This chapter has described the first step toward characterizing small tumors with acoustic angiography by using ROIs to delineate different tissue of interest (tumor, adjacent, distal, etc.). The use of growing ROIs for spatial analysis revealed that abnormal tortuosity extends beyond the border of the tumor itself, into the surrounding tissue located several millimeters from the boundary of a tumor. Thus, the extent of vasculature affected by a tumor is larger than the tumor itself. The implication of this finding is that if we consider abnormal vascular structure as the target we are trying to identify in order to identify a tumor, the region of abnormal vasculature is likely to be larger than the underlying tumor, giving us a larger “target” to identify. Figure 4.14, below, shows rendered vasculature from one of the tumors included in the preceding analysis with 2 different color-maps applied. The image on the left, in blue, shows distance from the tumor, where the tumor vasculature is shown by the lightest vessels and vessels in brighter shades of blue are located further from the tumor. This colormap required selection of a tumor ROI in order to define the tumor location and measure distances relative to the tumor. Therefore, while the image clearly shows the location
of the tumor and the surrounding vasculature, it requires an individual to define the tumor location and boundary. On the other hand, the image on the right is a color map indicating each vessel’s sum of angles metric tortuosity value. Vessels shown in yellow have the lowest tortuosity and brighter red vessels have higher sum of angles tortuosity. Since each vessel is color-coded based on its tortuosity metric, no user input is required beyond vessel segmentation. Despite the fact that the tumor location has not been defined in any way, the tumor is highlighted by a predominance of bright red, tortuous vasculature occurring in and around this tumor. Therefore, tortuosity may be valuable for detecting tumors, in addition to the characterizing their structure as was described in this chapter.
Figure 4.14: Segmented vessels are rendered and displayed in a color-map showing dis- tance from the tumor margin (blue) and sum of angles tortuosity (red).
The panel on the right is reprinted, with permission, from SR Rao*, SE Shelton*, PA Dayton. The fingerprint of cancer extends beyond solid tumor boundaries: assess- ment with a novel ultrasound imaging approach. IEEE Transactions on Biomedical Engineering, Vol. 63, No. 5, p. 1082-1086 (2016).
tumor or normal without first defining a tumor location or region of interest. Instead, classification will be based upon on only the vascular structure of the tumor and sur- rounding tissue. Chapter 5 uses a reader study approach to test visual classification, and chapter 6 combines the tortuosity metrics described in this chapter with vascular density using a clustering approach.
CHAPTER 5
QUALITATIVE ANALYSIS OF IMAGES: A READER STUDY APPROACH
5.1 OVERVIEW
After considering the tortuosity results generated from analysis on tumor ROIs and the spatial analysis of tissue surrounding tumors described in the previous chapter 4, the remaining question is whether an entire image can be classified as containing a tumor or as a normal image from a wild-type mouse. This goal of tumor detection or binary classification is distinct from previous analysis as its goal is predictive in nature. Given an angiography image, can we determine if it contains a tumor? This is a more challenging objective than statistical descriptions and comparisons between tumor ROIs and controls because we must develop an algorithm to allow us to label an image rather than comparing two pre-labeled groups. Therefore, this classification problem was approached in two ways, using visual assessment by trained readers, and using quantitative metrics derived from the vessel segmentations. The visual assessment includes a reader study completed by 7 readers who ranked the likelihood of belonging to the tumor class on a 1-6 scale. The quantitative analysis uses clustering analysis to integrate spatial information with multiple metrics of tortuosity and is described in chapter 6. Regardless of the approach, the image classification problem is much more akin to clinical detection and diagnosis of disease. For imaging technology to become clinically relevant, it must provide usefulpredicitve information, and not be restricted
topost hoc comparisons between groups.