Multimodal Medical Images Based Segmentation And Analysis Studies For Pituitary Adenoma

The visual disorders caused by pituitary tumors,such as visual field defects,vision acuity deterioration,color vision decrease,etc.,are treated by relieving the compression or injury to optic chiasm through surgical treatment,radiotherapy,and chemotherapy in clinical practices.Because the visual recovery situation varies from person to person,the evaluation and prognosis on the individual visual recovery before the treatment have become common issues of concern for neurosurgeons when they make the protocol or answer the patient’s questions.Based on literature research and meta-analyses,this dissertation validates the importance of the pituitary tumor volume,the effect of pituitary adenoma on the optic chiasm,and retinal measurement parameters to the prediction of post-treatment visual function recovery in pituitary adenoma patients.The research utilized the image segmentation and quantitative analysis methods by introducing multimodal clinical images,which included magnetic resonance(MR)and optical coherence tomography(OCT),to predict visual function recovery in patients with pituitary tumor.The approach in this dissertation provides technical support for the study of the relationship between the quantitative measurements from pituitary tumor images and the clinical prognosis indicator,serving new methods and new techniques to improve the diagnosis and treatment of pituitary tumor.The work and innovations of this dissertation are described as follows:1.A new three-dimensional(3D)interactive image segmentation method named randomwalk and graph cut based deformable model(RGCDM)was proposed to effectivelysegment the pituitary adenoma from the T1-weighted(T1W)MR images of thepatient’s brain.MR image segmentation experiments were performed to verify theeffectiveness and accuracy of the algorithm in the clinical analysis of pituitary tumor,which was superior to other related algorithms.2.A combined random walk algorithm was proposed to segment the optic chiasm fromT1W and T2-weighted(T2W)brain MR images interactively and efficiently.The 3Drending of the optic chiasm segmentation was combined with the pituitary adenomasegmentation obtained by the RGCDM method.It was convenient to observe therelative position between the pituitary adenoma and optic chiasm,as well as todetermine the squeezing situation of the optic chiasm.MR image segmentationexperiments were performed to verify the effectiveness and accuracy of the algorithmin the clinical analysis of optic chiasm.3.The retinal OCT image processing and quantitative analysis were investigated.Basedon the retinal spectral domain OCT(SD-OCT)image segmentation algorithm,theretina was divided into ten independent layers,followed by the extraction of themorphological features and light reflection characteristics of retinal OCT images.Addition to the measurement of the retinal layer thickness,the retinal OCT image lightreflection was innovatively proposed,paving a new way for the prediction of visualfunction recovery for pituitary tumor patients.In comparison with the quantitativemeasurements of a control group,it was concluded that the RNFL thickness hadsignificant difference between patients with pituitary tumor and normal people,andlight reflective characteristics had limited diagnostic ability,therefore furtherexplorations are needed in the future.