Computer-aided Prognosis Of Breast Invasive Ductal Cancer

Breast cancer(BC)is the most commonly diagnosed malignant tumor among females,and it seriously affects women's physical and mental health.The latest data from the China Health Statistics Annuals shows that BC ranks the first place of common female cancer incidence rate in China.What's worse is that the incidence of BC in our country increases year by year.The newly diagnosed cases were majorly early breast cancer among youth women.Improve the diagnostic rate of early breast cancer and the curative effect is the key subjects in tumor prevention and treatment.Breast invasive ductal carcinoma(IDC)is the majority type(60-80%)of BC.Thus,we should focus on tailoring therapies to improve IDC patients' living qualities and prolong their living time.According to Precision Medicine,tailoring therapies highly rely on accurate evaluation of patient's prognosis and therapy response.Though molecular taxonomy has been demonstrated to provide more accurate prognostic information for BC.But most molecular signatures are specially designed for ER(+),HER2(-)patients.And their actual role in clinical practice is limited by their availability.Thus,BC management relies largely on inexpensive and routine morphological assessment of HE images coupled with biomarker semi-quantification of IHC images.Histological features that are used to evaluate tissue architecture and cell morphology in HE images are key to pathology prognosis.Pathologists,the doctors' doctors play a critical role in BC treatment decision-making.However,HE image assessment via manual analysis has remained experience-based qualitative for over 100 years.Manual analysis always causes intra-or inter-observers variation even for experienced pathologists,which ultimately results in inaccurate evaluation.In this study,in order to obtain accurate prognostic information from HE histopathology images for IDC,we adopted a computer-aided image analysis method to quantify morphologic features.Since tumor nests(TNs)are the main functional units in breast cacner invasion,the prognostic value of TNs features becomes the emphases attention.The study of this paper is divided into two parts:Part One:Two step segmentation of hematoxylin-eosin stained histopathology images of breast invasive ductal carcinomaObjective In order to evaluate hematoxylin-eosin(HE)histopathology images quantitatively,this work proposes a computational method to automatically segment HE histopathology images of breast invasive ductal carcinoma(IDC).Method 1,180 digital HE images from 230 IDC patients were acquired.First,we proposed a pixel-wise color features and Haralick features based support vector machine(SVM)classifier for tumor nests from stroma segmentation.Then a method based on the marker-controlled watershed was adopted for cell nucleus segmentation.Pathologist annotations were used as judge standard to evaluate the effectiveness of our computational method.Result Pixel-wise features based SVM classifier segmented tumor nest from stroma with 87.1%accuracy,80.2%precision,81.0%sensitivity,and 89.9%specificity.And marker-controlled watershed can realize cell nucleus segmentation,especially for cell cluster.Conclusion Pixel-wise features based SVM classifier combined with marker-controlled watershed can segment breast invasive ductal carcinoma HE histopathology images automatically.Part Two:Computer-aided prognosis for breast invasive ductal carcinoma:based on quantitatively analyze of hematoxylin-eosin stained histopathology imagesObjective Computer-aided image analysis can help quantify morphologic features of hematoxylin-eosin(HE)histopathology images and provide accurate prognostic information for breast invasive ductal carcinoma(IDC).Method We performed a computer-aided prognosis(CAP)workflow on 1,150 HE images from 230 IDC patients.The pixel-wise SVM classifier was used to segment tumor nests(TNs)from stroma,and the marker-controlled watershed was used for nuclei segmentation.Multiple level features were extracted for TNs,stroma,and cells.Kaplan-Meier survival analysis and Cox proportional hazards regression model were used to evaluate the prognostic value of features.Results 730 features were extracted,and 12 features identified by Kaplan-Meier analysis were significantly associated with 8-year disease free survival(8-DFS)(p<0.05 for all).Moreover,4 features including TNs cell density(HR=1.625[95%CI:1.177-2.244],p=0.002),stromal cell structure feature(HR=1.596[95%CI:1.142-2.229],p=0.015),TNs feature(HR=1.327[95%CI:1.001-1.759],p=0.032),and TNs cell nuclei feature(HR=0.729[95%CI:0.537-0.989],p=0.045)were identified by Cox proportional hazards model to be new independent prognostic factors.Conclusions The results indicated that computer-aided image analysis could quantify HE histopathology images and dig out prognostic information for IDC.The TNs cell density,and stromal cell structure feature,TNs feature,and TNs cell nuclei feature could be serviced as new prognostic factors for IDC.