Intelliqent Scanning And Intelligent Screening

Obstetrics ultrasound examination and gynecology cytology examination are common used examination techniques in obstetrics and gynecology. Among them, the key task of prenatal ultrasound diagnosis is to obtain the image of standard plane of interested anatomies and to perform biometric measurements. However, the accom-plishments of these tasks mainly depend on the experiences and skils of the radiolo-gist. The standard for screening cervical cytology is for a cytoscreener to manually search across an entire smear for abnormal cells using a conventional microscope. The computer-assisted cytological test (CCT) can automatically select abnormal cells, which allows targeted reading by cytoscreeners. However, the CCT has not been used widly in clinic at present.To reduce the user dependency of prenatal ultrasound diagnosis results, we pro-pose an idea of Intelligent Ultrasound Scanning (IUS), which aims at automatically locating of the standard plane of early gestational sac (SPGS) and performing bio-metric measurements based on image analysis methods. To promote the clinical ap-plication of CCT, we propose to perform Intelligent Cytology Screening (ICS) on hematoxylin and eosin (H&E) stained manual liquid-based cytology (MLBC) slides. The aim is to verify the performance of ICS system is comparable to that of CCT with proprietary stained automated LBC (ALBC) slides.The proposed intelligent ultrasound scanning (IUS) algorithm framework con-sists of three steps. First, the GS is efficiently and precisely located in each ultrasound frame by exploiting a coarse-to-fine detection scheme based on the training of two cascade AdaBoost classifiers. Next, the SPGS are automatically selected by eliminat-ing false positives. This is accomplished using local context information based on the relative position of anatomies in the image sequence. Finally, a database-guided mul-tiscale normalized cuts algorithm is proposed to generate the initial contour of the GS, based on which the GS is automatically segmented for measurement by a modified snake model. This system was validated on31ultrasound videos involving31preg- nant volunteers. The differences between system performance and radiologist perfor-mance with respect to SPGS selection and length and depth (diameter) measurements are7.5%±5.0%,5.5%±5.2%and6.5%±4.6%, respectively. Additional validations prove that the IUS precision is in the range of inter-observer variability. We conclude that IUS of the GS from2D real-time ultrasound is a practical, reproducible and relia-ble approach. In addition, after some extension, the proposed IUS algorithm frame-work has the potential to be applied to the scanning task of other fetal anatomies.Our research on ICS sytem mainly focuses on the automatic cell segmentation and classification. Cell segmentation:1) the proposed method combining the A*channel and the hybrid multi-way cuts can accurately segment the cytoplasm on H&E staining images;2) the proposed Local Adaptive Graph Cuts algorithm can improve the segmentation precision of cell nuclei, especially abnormal cell nuclei; and3) the combination of two concave-based splitting algorithms can effectively split overlap-ping cell nuclei. Cell classification:1) we design a supervised learning-based cyto-logical classification framework;2) the newly incorporated features including rough-ness indexes and local binary pattern (LBP) means can effectively eliminate artifacts and normal cells;3) the sensitivity of classifiers to discriminating abnormal cells is increased by using feature preprocessing technique; and4) the specificity of normal cell recognition is improved by incorporating cell nuclei area context information and by using approximate cytoplasm features to exclude difficult negatives. Cyto-pathologists can get a sensitivity of88.1%and a specificity of100%when using the ICS system for prescreening on43smears (21abnormal and22normal). Moreover, for rescreening, beginner can find out8.3%suspected abnormal smears from those diagnosed as normal by cytopathologists, of which there is25%false negative rate. Clinical trials show that both the sensitivity and the specificity of the ICS system are satisfyingly high, thus proving the feasibility of the ICS for MLBC with H&E stain-ing. Therefore, the current research may lead to wider application of the CCT tech-nique.