Active Shape Modelling in the Prediction of Hip Fracture - Research Proposal Example

CHAPTER FOUR DATA ANALYSIS 4.0 Introduction This section of the project is dedicated to undertaking a critical analysis of the data that have been gathered so far. It would be realized that there have been two major forms of data collected, which are quantitative data and qualitative data. To this end, there shall be both quantitative analysis and qualitative analysis of data. Qualitative data were collected in the form of review of secondary data whiles quantitative data were collected in the form of primary data, which were collected by setting two major respondents thus control and cases. In order to ensure that the discussion and analysis are well organized and structured, they have been done in the form of themes and sub-sections as they related to the data collection. 4.1 Presentation of Results Running through the results that were obtained as a result of the observations and series of experiments undertaken by the researcher, there are some key patterns that are noticed. Some of these major patterns are presented as followers 4.1.1Decrease in All Runs It would be noticed that prior to the actual implementation of the use of the active shape modeling, there were series of runs that were conducted. The runs rate of progression in these runs were recorded by using three different media as a way of consolidating results that were obtained. In effect, results of mean, median and standard deviation were all tested. There were also two major aspects of the runs, which are given as point-to-point and point-to-line. The results obtained from all these basis pointed to one major phenomenon and that there was a shape decrease in each of the runs that were recorded (Baluja et al. 1995). The interpretation given to this decrease in all the runs is given in the next section. 4.1.2 KL Grading Score Other core patterns of the whole research were also obtained from the KL grading score that was undertaken. The grading was done with reference to the two runs that were done earlier, which has been presented above. These two runs were the point-to-point and point-to-line. After careful observation, it was realized that as many as 98% of the KL grading fell within the range of 1 KL grade. Of the grades also, 46% were recorded to be identical with a high value of 91% of the KL grade that changed increasing their KL score. On this score also, interpretations are given in the next major section of the chapter. 4.1.3 Distribution of KL Grade in Control and Case Groups Earlier at the beginning of the research, there had been two major sample groups drawn. These were basically the control group and the case group. The groupings were necessary to ensure that there existed some basis of comparing results that were obtained from the two groups to find the effectiveness of the intended intervention. In consonance with this, a separated attention was given to KL grades recorded in the control and case groups. An important pattern that was recorded was that, there continued to be 82% of control having KL grades of 0 or 1. On the part of the case group however, the pattern was that there was even distribution of KL grades. For example two thirds of the respondents had KL of either 0 or 1. Among this group however, ¼ had grade 1. 4.1.4 Distribution of Modes Series of modes were tested for study in their pattern of distribution. In the by and large, the pattern recorded gave a realization that not all the modes showed exactly the same discretion of distribution. For instance modes 6, 8 and 12 were not normally distributed. What this meant was that there was the need for a separated system of testing, which was done by using Mann-whitey test. At the end of the new test, a new pattern that was detected was that modes 6 and 8 were not significantly different; meaning no further changes was still recorded. However, mode 12 cases were significantly higher than those of control. This pattern is another indication of the differences that exists between the case groups as compared to the control group (Bland and Altman, 2006). 4.2 Interpretation and Analysis of Results 4.2.1 Resulting Predictions Based on the patterns presented above, there are some key resulting predictions that can be hypothesized. In the first place, the decreasing results in the runs that were obtained give a resulting prediction that there is a highly reliable and clear range of consistencies in the predictions of possible hip fractures and other forms of irregularities with the bone. If for nothing at all, there will not be ambiguous, fluctuating and unstable results. Furthermore, this is a major indication that the operator of the system and thus the researcher was gaining more and more improvement in the application of the system through the training process (Christopher and Burges, 2008). This was another important milestone in solidifying the authenticity of the eventual result. With regards to the KL grade scores, the resulting prediction is that upon the application of the intervention, all respondents will not react in exactly the same manner towards the KL system. As with all hip prediction systems, it is very important that there can be a highly close relationship between the results to be obtained for as many groups of respondents and users as possible (Cristinacce, 2004). However, the research has revealed that on a scale of 0 to 2, even though a larger portion of the grade is likely to be 1, this was not going to be consistent with all respondents. In effect, the need to identify a new system that seeks to focus on the core competence of scaling down the differences in grade score is of major importance (Hothorn, 2007). 4.2.2 Exceptions to the Patterns Already, the results give certain indications that the patterns that were recorded in the processes will not always remain the same. For instance it would be noted that the control and case groups gave out different patterns. Invariably, there is a major mechanical implication to this whereby in the actual application of the system in the identification of hip and bone related predictions, users who undergo certain interventional routines may possibly display different results from other ordinary people (Cristinacce and Cootes, 2003). Even apart from these, other variables such age, sex and body mass index may also affect the patterns and thus cause key differences in the results of predictions. In a recent study for instance, it was identified that age is one of the most outstanding factors that can cause deviations and exceptions in hip fraction prediction as the age of a person largely determines the tenderness and penetrative range of a born (Stephan Al-Zubi, 2004). 4.2.3 Mechanisms underlying Patterns The aim of this research, which was to compare active shape modeling to hip morphometry in the prediction of hip fracture could possibly not be completed without reviewing the existence of Osteoarthritis, which is highly associated with the use of both morphometry and the active shape model in the United States. The secondary data reviewed disclosed that Osteoarthritis “is one of the most common ailments of the musculoskeletal system and that it is commonly known as degenerative joint disease or ‘wear and tear’ arthritis, OA (Murray & Lopez, 1997). The researcher therefore undertook a data collection process to identify the rate of OA risk factor among United States citizens. The data collection revealed that people aged 65 and over were at the worse risk of infection. Unfortunately, Murray & Lopez (1997) estimates that “by the year 2020 it is estimated that 71% of developing nations will be aged over 65 years.” This means that there is an urgent need to look for an intervention that will improve the predictive nature of hip fraction; not just in the United States but in the world over. The National Health Service gives statistics on the prevalence rate of OA risk that demands immediate attention. This is because according to the service, “46 million Americans above the age of 25 are affected by OA”. To policy administrators and health reform enforcers, the meaning of this statistics is that there should be a pragmatic approach in identifying the existence of hip fractions at a very early stage of the disease. But before this can be possible, it is important to have a reliably system that can predict hip fractions at an early stage. This is one important concern that informed the present research to ascertain the rate of accuracy in the prediction of hip fraction by using active shape model. 4.2.4 Agreement or Disagreement with previous work The studies has greatly revealed the merits that comes with the use of the active shape modeling system in the prediction of hip and other forms of fraction. Especially in comparison to the Hip Morphometry, it would be seen that the active shape modeling is a more advanced and technology based system that makes the work of the professional easier and more accurate. Apart from the ordinary role in the identification of hip fractions, literature reviewed showed that the use of active shape model paves the way for automatically searching images for new instances of that object (Grimson, 1991). This is indeed a benefit that other means of fraction prediction do not provide. With such roles taken care of, the practitioner then has enough time to concentrate on more manual oriented parts of his work, thereby achieving general efficiency (Edwards, Taylor and Cootes, 2008). 4.3 General Discussion of Results A training section was instituted for the perfection on the use of the software that comes with the active shape modeling. This paved way for there to be a series of run tests conducted on the software. This was done to achieve the rate and levels of repeatability of the software. It would be noted that “Repeatability is defined as the set of variations when the same object under the same set of conditions and using the same instruments is measured by an individual” (Gregory et al. OP International) There were therefore the indication of conditions that functioned in the same manner. Out of these, the mean, median and standard deviation of all the runs that were carried out were measured. There were two major forms of the run, which were point-to-point and point-to-line. Interestingly, results showed a decrease in each of the runs as the runs progressed. Quantitatively, this is a perfect indication that the software keeps recording consistency and in effect, an improvement in its function and use. The active shape modeling could therefore be judged as a productive intervention for predicting the hip fracture that exists in the bone. Several shape models were sampled for analysis. To undertake a widely reliable result, there were as many as 15 models. The general idea was to examine the normality of the various shape models that were sampled using the Shapiro-Wilks test. Consequently, parameters were set to from which the range or size of effect of the shape models will be determined as either being small, medium or large. The parameters used were the mean, which were to be interpreted as 0.0 -0.2 for small effect, > 0.2 -0.5for medium effect and >0.5 for larger effect (Cohen, J. 1988). Again, there was another parameter set, which was to compare the size of effect between the control and case. This was necessary to distinguish the differences that exist between these two components. From table 3.9, there is an indication that when emphasis was placed on mode 12, the average value that was recorded for both the case and the control was 0.2. This is an indication that indeed, the size of effect is very minimal; thus small. 4.3.1 Relationship between Results and original Questions The KL grading is one of a system that has largely been used in several circumstances where there was the need to locate the existence of hip joints. However, the working rate of this system; when it is made to function in isolation has often been criticized by experts as not being very effective and efficient. It is for this reason that there was the need to locate and identify a suitable component that could speed up the predictive nature of the system. This led to the use of the active shape modeling system. When the assessment was done on the impact of the active shape model in adding to the predictive ability of the KL grading, it was discovered that when forward stepwise method (Ward’s Statistic) was used, there were actually very significant improvement in the predictive ability of the KL grading. This is therefore a major indication to the achievement of the ultimate aim of the research, which was to compare active shape modeling to hip morphometry in the prediction of hip fracture (Friedman, 2001). CHAPTER 5 CONCLUSION The conclusiveness of the reliability and validity of the data that were collected were very much dependent on the independence of the independent samples t-test that was carried out. It was expected that there the variables that were collected from the cases and control would both match. In an empirical study of this nature, the distinction between cases and control ought to be close and highly similar. In this regard, it can be said that the expected results were achieved as the table 3.6 shows that there was a great match between the variables from both the control and cases. Typical examples of these could be given with independent variables like age, body mass index, average height and average weight. As indicated earlier, with research work of this nature where control and cases samples are set, the researcher always has a goal of proving that the application of a number of interventions on the cases could have an effect that may be absent in the control sample. For this reason, there should not be any predetermine factors that result from unmatched variables causing such differences in results. The independence of the samples t-test was therefore perfectly done. Results from the other chapters and the discussion in the previous chapter can be used to draw a number of conclusions as far as the aim of the research, which was to compare the active shape modeling compared to hip morphometry in the prediction of hip fracture, is concerned. These conclusions are made in four major themes. First, the data collection and discussion brings out the core need for there to be an innovative system to track the existence of hip fraction in humans. This is because of the high prevalence rate of OA in most citizens of the United States and other parts of the world. Indeed, with the scaring estimates that “by the year 2020 it is estimated that 71% of developing nations will be aged over 65 years (Murray & Lopez, 1997), there could not be nothing more done than to finding a system that would ensure that the prediction of the existence of a hip fracture is done at an early stage so that every needed medical attention can be given before things degenerate into complexities. Secondly, it can be concluded that the sue of active shape modeling has by results from the research proven to me the most effective way of increasing the chances and accuracy with prediction of the existence of hip fracture in humans. This is because of the high rates of prediction scored through the various mean marks recorded. Invariably, the health fraternity is advised on the need to stick to such technology based and innovation oriented means of predicting the existence of hip fracture. This assertion is made against the backdrop that unlike other systems of identifying the existence of hip fractures, the active shape modeling does away with most manual manipulations and therefore makes the rate of accuracy and speed of work as perfectly as possible. Indeed, with the large numbers of potential sufferers of OA lining up the hospitals each day, the only way to remedy the situation is to have an equally highly efficient system and that is the active shape modeling. Specifically on the research approach used, it would be concluded that, the data collection and research implementation process were highly valid and reliable. This gives the present research work a scientific research basis, which is supposed to be empirical (Hewson, 2007). This conclusion is drawn from results obtained from the closeness in the match of variables that were collected and sampled as cases and control. Clearly, because of the closeness in the values received, there could be no way that the results gathered on the effectiveness of the active shape modeling could be attributed to manipulations from the researcher as a means of fielding biased variables for the cases and control sample groups respectively (Galassi et al., 2004). The final conclusion is an advocacy that the fact that there has been a strong case through this research of the effectiveness of the active shape modeling should not be a means for stopping all forms of medical interventions that are targeted and channeled towards the prevention of cases of OA and all other forms of fractures in healthy people. This is because as the adage goes, prevention would always be better than cure. Therefore, it is very important that medical experts continue to look for interventions and innovative ways of preventing the occurrences of OAs in patients. As much as possible, the use of the active shape modeling should be fallen upon as a last resort to handling very complicated cases of the existence of hip fractures. REFERENCE LIST Baluja T, Kanade H, Poggio A, Rowley Y. A, and Sung S. K. 1995. CMU Frontal Face Images. Carnegie Mellon University, Robotics Institute, vasc.ri.cmu.edu/idb/html/face/frontal_images. Bland J.M. and Altman D.G. 2006, Statistical Methods for Assessing Agreement between Two Methods of Clinical Measurement. Lancet, i, 307{310, www.users.york.ac.uk/~mb55/meas/ba.htm. Christopher J. C. and Burges. 1998, A Tutorial on Support Vector Machines for Pattern Recognition. Data Mining and Knowledge Discovery 2:121{167, Cristinacce D. Automatic Detection of Facial Features in Grey Scale Images (Doctoral Thesis). University of Manchester (Faculty of Medicine, Dentistry, Nursing and Pharmacy), 2004. mimban.smb.man.ac.uk/publications/index.php. Cristinacce D.and Cootes T. A Comparison of two Real-Time Face Detection Methods. 4th IEEE International Workshop on Performance Evaluation of Tracking and Surveillance, 2003. mimban.smb.man.ac.uk/publications/index.php. Edwards G., Taylor C. J. and Cootes T. F. 2008, Interpreting Face Images Using Active Appearance Models. 3rd International Conference on Automatic Face and Gesture Recognition 1998, pages 300{305, Japan, Friedman J. H. Multivariate Adaptive Regression Splines (with discussion). Annals of Statistics 19, 1, 2001. www-stat.stanford.edu/~jhf. Galassi et al. 2006.GNU Scienti_c Library Reference Manual. GNU, www.gnu.org/software/gsl. Hewson P. CRAN Task View: Multivariate Statistics. R language online document, 2007. http://cran.r-project.org/src/contrib/Views/Multivariate. html. Hjelm E and Boon K. L. Face Detection: A Survey. Computer Vision and Image Understanding, 83(3), pages 236{274, 2001. Hothorn T. (maintainer) CRAN Task View: Machine Learning and Statistical Learning. R language online document, 2007. http://cran.r-project.org/src/contrib/Views/MachineLearning.html. Stephan Al-Zubi. Active Shape Structural Model (Doctoral Thesis). 2004, Otto-von-Guericke University of Magdeburg, citeseer.ist.psu.edu/732672.html. Read More