METODI PER L’ANALISI DELLA TEXTURE DI REGIONI DI INTERESSE (ROIs) DI FORMA IRREGOLARE PER IMMAGINI BIOMEDICHE

This thesis presents algorithms for the assessment, in irregular gray scale image ROIs, of the fractal dimension (FD) and indices aimed at the characterization of the related histograms. The tools were tested by their applications in the two studies. The first study shows the preliminary results obtained in the experimentation of a new method for liver steatosis quantification by means of ultrasound image analysis. This method is based on hepato-renal ratio (HRR) assessment. HRR was assessed as the ratio between the mean intensities of pixels belonging to two ROIs located respectively in the liver and the kidney. In the method outlined here, the operator selects two ROIs by “freehand”(fh ROIs).The ROIs were subsequently pre-processed, via “ROI normalization”, to make the two areas more comparable, and smoothing out the related normalized histograms. Then HRRs were calculated on the basis of the histograms obtained from pre-processing(PP). In order to test the efficacy of the PP, the procedure was repeated by a second operator, using elliptical and/or circular ROIs(e/c ROIs). The dataset comprised 98 images of 32 patients(for 23 patients more than one scan was performed). The efficacy of the technique presented was tested by comparison with a method developed on the basis of previous works. The HRRs obtained with ROIs pre-processing did not differ significantly from those of the reference method (for the fh ROIs and e/c ROIs the p-values were 0.52 and 0.14 respectively). A similar result was obtained by comparing the HRRs derived from the PP of the e/c ROIs and those resulting from the PP of fh ROIs (p- value=0.46).The method for HRR assessment presented here (on the basis of the PP of fh ROIs) gave a smaller inter-scan variability than the reference method (p-value<0.001). Preliminary results seem to confirm that the proposed method is reproducible and has clinical efficacy comparable to the reference method. The second study aimed at the quantitative assessment of the ultrasound alterations of the internal anal sphincter(IAS), associated with aging. 62 healthy subjects aged between 19 and 87 underwent 3D endoanal ultrasonography. The medial part of the IAS was characterized by assessments of the volume, median voxel intensity (MI) and FD. The age of the subjects under analysis was classified as a three-level variable(19-59, 60-69 and 70-87years). The associations between age and IAS characteristics were evaluated by a path analysis(PA).The MI and the FD were included in the PA as potential mediators of the effect of age on the volume only if they showed an increasing trend according to aging by descriptive analysis. The association between each pair of variables was summarized using the ratio of expected values(REV: 1=no association, <1=negative association, >1=positive association).In descriptive analysis, unlike for volume and MI, the FD did not show an increasing trend according to aging, therefore it was excluded from PA. In the PA, the MI showed a statistically significant difference between older and younger subjects only(70-87vs.19-59 years, REV= 1.21, p-value=0.048). Furthermore solely a direct positive effect of age on volume was observed(60-69 vs.19-59 years, REV=1.21, p-value=0.070 and 70-87vs.19-59 years, REV=1.32, p-value=0.015).The MI seems to be an appropriate measure of echotexture alterations in the IAS related to aging.The findings obtained appear not to support the hypothesis that there is an association between the age-related physiological process of fibrosis and a volume increase with age. For both works presented, the results obtained and/or related interpretations require verification by further investigation via studies aimed at the validation of ultrasound assessments by histological findings, analyzing a large sample of subjects. These studies are examples of how texture analysis can be used to provide new tools able to perform objective and quantitative assessments for clinical purpose