腫瘤影像特徵之量化與效力分析以甲狀腺與乳房腫瘤超音波影像及肝癌電腦斷層掃描影像為例

Tumor Feature Quantification and Performance Analysis and Its Applications to Thyroid and Breast Tumor Sonography and Hepatocellular Carcinoma Computerized Tomography

林彥君(Yen-Chun Lin)

指導教授 ： 陳正剛

國立臺灣大學/工學院/工業工程學研究所/碩士(2013年)

https://doi.org/10.6342/NTU.2013.10548

摘要

醫學影像為有效檢查早期腫瘤發生之非侵入性篩檢方式之一，醫生藉由觀察醫學影像之特性提出持續追蹤該腫瘤或進一步接受細胞學檢查之建議。本研究目標乃利用電腦客觀定量的分析醫學腫瘤影像之特徵。傳統解釋醫學影像特徵乃主觀憑藉著自身經驗以及觀察做出臨床診斷。因此本研究將著重在建立適當之醫學腫瘤影像的客觀量化指標以降低人為主觀判斷的差異。利用量化的客觀指標使觀察者自身及觀察者之間的差異最小化。

本研究探討主題其一為建立甲狀腺腫瘤超音波影像之量化指標，例如腫瘤邊緣模糊性。已知具有邊緣模糊的甲狀腺腫瘤有較高的風險為惡性腫瘤。然而目前為止尚未有文獻提出合適的邊緣模糊性量化指標，因此本研究開發新的邊緣模糊性量化指標，幫助臨床醫師診斷。

本研究其二為探討量化的腫瘤影像特徵指標是否能夠幫助乳癌患者的預後。針對電腦輔助診斷乳房腫瘤超音波影像雖有許多文獻提出量化指標，但乳房遠端轉移與非遠端轉移之惡性腫瘤在超音波影像上之量化指標差異性卻尚未被探討與研究。可能發生遠端轉移的病人在預後若能獲的相對應的治療與照顧，將能降低復發得情況並進一步的延長患者生存的機會。

探討主題其三為建立肝癌 (Hepatocellular Carcinoma, HCC) 電腦斷層掃描 (Computerized Tomography, CT) 影像之量化指標。CT 是診斷肝癌和預後最主要的工具之一，然而 CT 影像的腫瘤特徵與其復發率之間的相關性卻鮮少被提及與研究。因此本研究發展 CT 影像的量化指標，辨別影響肝癌復發的重要因子。

本研究使用臺大醫院所提供之臨床數據以及醫療影像。甲狀腺腫瘤邊緣模糊性與乳癌預後提出的指標將利用t-檢定 (Student's t-test) 的 p-value，並利用接收者操作特徵曲線 (Receiver Operative Characteristic Curve, ROCCurve) 做進一步的驗證。對肝癌影像特徵所提出之量化指標，本研究使用存活分析以及 Cox 比例風險模型Cox Proportional Hazards Model(Cox Regression) 進行績效之驗證。結果顯示顯著之指標將有助於診斷腫瘤邊緣模糊性、乳癌遠端轉移的預後以及肝癌的復發。

關鍵字

醫學超音波影像 ； 電腦斷層掃描影像 ； 特性量化 ； ROC 績效 ； 腫瘤臨床特性

並列摘要

The medical imaging is one of the most effective non-invasive screening tools for early tumor diagnosis. Based on the characteristics observed on the medical images, cliniciansmake recommendations for patientsto keep track of the tumors or undergo further cytology tests.This research aims on computerized quantification of themedical image characteristics. The traditional interpretation of medicalimages is mostly subjective and highly dependent on the medical staffs’ experience and judgment. This research focuses on how to establish objective quantified indicatorsof the medical image characteristics.By using quantitative and objective indicators, the intra- and inter-observer variability can be minimized if not eliminated.

The first part of this research is to quantify an important sonographic characteristic of thyroid tumors, i.e. the margin blurriness. It is known that a tumor with a blurred margin has a higher risk of being malignant cancer. However, up to now, thereis no proper indicator to quantify the blurriness of the margin in the literature. Therefore, this study is to develop novel indicators to objectively quantify how blurred the tumor margin is and help clinicians make their recommendations.

The second part of this research is to investigate whether the quantified indicators of the image characteristics can help the prognosis of breast cancers. Though there have been plenty of studies in computer-assisted diagnosis of breast cancers using quantified image characteristics, how the indicators help predict distant malignant metastases has not been studied and mentioned in the literature. The prognosis of distant metastases with corresponding treatmentsand appropriate cares may reduce the probability of recurrence and further extend patients’ survival rate.

The third part is to develop quantified indicators for computerized tomography (CT) images of hepatocellular carcinoma (HCC). CT is one of the most important diagnostic and prognostic tools of HCC. However, correlation between CT image characteristics of HCC and the recurrence rate is rarely mentioned in the literature. Therefore, this study developed quantitative indicators of the HCC CT images to help identify the significant factors that have impacts on the HCC recurrence.

The research uses clinical data and medical images provided by National Taiwan University Hospital (NTUH).The developed indicators for the tumor margin and the breast cancer prognosis are analyzed with student’s t test and its p-value and further validated by the receiver operating characteristic (ROC) curves to validate the performance. For the HCC　CT image characteristics, survival analysis and Cox proportional hazard model are used to validate the performance of the proposed indicators. The results show that the significant indicators can be found in the detecting blurred margin and in prognosis of breast cancer metastases and HCC recurrence.

並列關鍵字

Ultrasound Image ； CT ； Feature Quantification ； ROC Performance ； Clinical Features

參考文獻

[1] Horsch, K., Giger, M. L., Venta, L. A., & Vyborny, C. J. (2002). Computerized diagnosis of breast lesions on ultrasound. Medical Physics, 29, 157.

[2] Giger, M. L., Al-Hallaq, H., Huo, Z., Moran, C., Wolverton, D. E., Chan, C. W., & Zhong, W. (1999). Computerized analysis of lesions in US images of the breast. Academic radiology, 6(11), 665-674.

[3] Moon, W. J., Jung, S. L., Lee, J. H., Na, D. G., Baek, J. H., Lee, Y. H., ... & Lee, D. H. (2008). Benign and Malignant Thyroid Nodules: US Differentiation—Multicenter Retrospective Study1. Radiology, 247(3), 762-770.

[4] Papini, E., Guglielmi, R., Bianchini, A., Crescenzi, A., Taccogna, S., Nardi, F., ...& Pacella, C. M. (2002). Risk of malignancy in nonpalpable thyroid nodules: predictive value of ultrasound and color-Doppler features. Journal of Clinical Endocrinology & Metabolism, 87(5), 1941-1946.

[5] Chou, C. T., Chen, R. C., Lee, C. W., Ko, C. J., Wu, H. K., & Chen, Y. L. (2012). Prediction of microvascular invasion of hepatocellular carcinoma by pre-operative CT imaging. British Journal of Radiology, 85(1014), 778-783.