Near real time confocal microscopy of Ex Vivo cervical tissue: detection of dysplasia

Recent studies have shown the ability of confocal microscopy to noninvasively image cells in vivo in real time. This ability to visualize nuclei in vivo shows the potential of confocal microscopy to dramatically improve the prevention, detection and therapy of epithelial cancers. More exciting is the potential to quantitatively measure nuclear morphometry providing a basis to automate the cancer detection process. This dissertation describes studies exploring this potential in ex vivo cervical tissue using acetic acid as a nuclear contrast agent. First the use of acetic acid was demonstrated to improved contrast in confocal images of cervical tissue sufficiently to allow segmentation. Segmentation is robust throughout the epithelium in most normal tissue and upper portions of tissue diagnosed with severe dysplasia. Based upon this segmentation, quantitative feature measurements were extracted from confocal images of cervical tissue in a pilot study to determine if the features would aide in the detection of dysplasia. Simultaneously, a qualitative review of confocal images was performed by untrained reviewers and compared with clinical colposcopic impressions, the standard clinical tool aiding in dysplasia detection. The sensitivity and specificity of both the qualitative (95% and 69%) and quantitative (100% and 91%) review were improved compared to colposcopic review (91% and 62%). Finally the ability of confocal microscopy to produce 3D images was explored as a further means to improve dysplasia detection. Based upon Beer’s equation for light attenuation, the scattering coefficient was extracted from 3D image sets of ex vivo cervical tissue and compared with histology from the same precancerous lesion. The results suggested a possible correlation between high scattering values and the presence of dysplasia. Quantitative 3D features were also extracted from 3D image sets and correlated with the presence of CIN 2/3. Increased separation between normal and CIN 2/3 biopsies was produced using the 3D features as compared to the 2D. More importantly, when additional information (scattering coefficient) is combined with the 2D features, the ability to distinguish between normal and CIN 2/3 is 100% accurate in this small sample set.