The quantitative and qualitative evaluation of ovarian carcinoma using multi-modality imaging approach

Abstract

Computed tomography (CT) is widely used to evaluate gynaecological disease. Incidental ovarian findings are common on CT; about 10.0 % of incidental ovarian masses are malignant. Ovarian carcinoma is a heterogeneous group of cancers that originates from the adnexal structures. The immediate determination of ovarian mass on CT could help patients avoid additional examinations. The early diagnosis of ovarian carcinoma before disease has metastasised could prolong patients’ survival. For patients confirmed with ovarian carcinoma, pre-operative CT is recommended to assess the disease dissemination. The differentiation of histological subtypes of ovarian carcinoma could help patients proceed to suitable management. Texture analysis is promising in quantitatively characterizing image patterns to reflect tumour heterogeneity. In this thesis, the ability of CT texture analysis in determining the nature of ovarian masses was explored. CT texture analysis could achieve an area under receiver operating characteristic curve (AUC) of 0.95 in differentiating benign ovarian mass from early-stage ovarian carcinoma using support vector machine (SVM); and an AUC of 0.91 for the determination of histological subtypes in ovarian carcinoma using random forest model. Treatment of ovarian carcinoma involves cytoreductive surgery in combination with chemotherapy. The use of neo-adjuvant chemotherapy (NACT) can increase the likelihood of optimal resection in a subset of patients, but there are no established criteria for the treatment response assessment after NACT. Herein, the ability to detect residual disease after NACT was compared between CT and positron emission tomography-computed tomography (PET-CT). Our study demonstrated that neither qualitative assessment of CT nor PET-CT could preclude a thorough surgical assessment for residual disease after NACT considering the low negative predictive value (NPV) of CT (50.0 %) and PET-CT (16.7 %). In final sections of the thesis, we introduced a semi-quantitative method to accurately segment functional tumour burden in patients with advanced and recurrent ovarian carcinoma using diffusion-weighted imaging (DWI). The DWI-derived functional tumour burden was highly correlated to surgical derived tumour burden assessed by peritoneal cancer index (PCI). High DWI-derived tumour burden was a negative predictor for surgical outcome and complexity. The proposed logistic regression model could achieve an accuracy of 92.9 % for the prediction of complete cytoreduction. The overall survival was longer in low tumour burden group than high tumour burden group (432.0 vs. 833.0 days, p = 0.021) with hazard ratio (HR) of 3.635. In summary, these studies showed the value of multi-modality imaging (CT, MRI, PET-CT) in the quantitative and qualitative evaluation of ovarian carcinoma, including diagnosis, NACT treatment response assessment and treatment stratification. Various analytical methods were applied. Nevertheless, the studies presented in this thesis were limited by small sample sizes. Furthermore, heterogeneity of the imaging data could induce research variability between centres. Thus, these results will require validations in large, harmonized dataset before clinical translation.