Desarrollo y validación de sistema CAD para la cuantificación de parámetros en el estudio de imágenes médicas de la columna vertebral

Resum

The misalignments of the spine can lead to limitations in the quality of life, especially idiopathic scoliosis in the adolescence (AIS) because of the secondary conditions that can cause in severe cases. The full spine frontal plane X-ray in standing position is the gold-standard imaging test for the diagnosis of AIS. Specific measurements on the X-ray are required, including axial vertebral rotation (AVR) and Cobb angle in order to correctly assess, diagnose, follow-up and treat AIS. Methods traditionally used to quantify these variables by hand introduce have bias in decision-making regarding treatment options. Technology development in computing and digital X-ray includes CAD software that allow simpler and more preside measuring on radiographs these images. Although there are currently commercial CAD systems that measure these clinically relevant variables in the X-rays of spine misalignment, they important limitations: lack of specificity; lack of adequate tools; long training or operating times may be required; may be not compatible with personal home computers; they do not allow functionality changes functionalities, such as implementing new customized tools for each clinician or researcher´s needs; high costs. The overall goal of this PhD thesis was to develop and validate a Computer Aided Quantification software (CADq) for use in research. This software quantified the required variables of the assessment of spine misalignments, and overcame the limitations described above. Its features are to be easier to use, and more precise and accurate than the traditional measurements by hand, as well as being applicable as research tool. Because the topic was very broad, it was decided to limit the work of the thesis to the measurement of the Cobb angle and vertebral rotation in idiopathic scoliosis. The CADq software developed has been registered under the name TraumaMeter, using C++ language and OpenCV machine vision libraries, which incorporates measurement tools for various variables, including AVR and Cobb angle. A dimensionless linear mathematical theoretic equation has been calculated that allows implementing the ability to measure AVR in any CAD system, with significant improvements over the non-instrumental Raimondi method. The quality of the AVR measurements obtained by implementing the developed mathematical equation in the TraumaMeter software was evaluated. Likewise, the non-instrumental measurement method was evaluated and compared to our calculations. Our implemented equation in TraumaMeter software obtained measurements 1.9 times more reliable and 1.72 times more valid than the non-instrumental method. The agreement between the two AVR measurement methods (software and non-instrumental) is almost maximum (ICC(2,1)=0.975). The validity and accuracy of Cobb angle measurements using TraumaMeter has also been assessed, which improves measurements by removing some sources of error intrinsic to the Cobb method, and reducing others such as imprecise selection of reference points of the terminal vertebrae. Likewise, the measures by hand were evaluated and compared to our software quantifications. The results showed that TraumaMeter generates more accurate and precise quantification than the measurement by hand. Besides, traditional method of measures (gold standard) and TraumaMeter method were almost equivalent, showing almost absolute agreement (ICC(2,1)>0.99). It has also been observed that non-experienced examiners can measure the Cobb angle with high validity and reliability using our software from the first time. In quantifications by hand, small differences in measures (around 0.5º) could represent real changes. Due to the high accuracy and precision of the TraumaMeter software when measuring vertebral rotation, it was also used to study whether there are statistically significant differences in the error of AVR measurements in X-rays, according to the type of vertebra. It was concluded that the error in the AVR measurements of the thoracic vertebrae is significantly greater than the error in the AVR measurements of the lumbar vertebrae (1.38º with respect to 0.38º). The reliability of the measurements of the thoracic and lumbar vertebrae was almost the same (1.88º compared to 1.83º).