A Learning Based Hierarchical Framework for Automatic Prostate Localization in CT Images

Abstract

Accurate localization of prostate in CT images plays an important role in image guided radiation therapy. However, it has two main challenges. The first challenge is due to the low contrast in CT images. The other challenge is due to the uncertainty of the existence of bowel gas. In this paper, a learning based hierarchical framework is proposed to address these two challenges. The main contributions of the proposed framework lie in the following aspects: (1) Anatomical features are extracted from input images, and the most salient features at distinctive image regions are selected to localize the prostate. Regions with salient features but irrelevant to prostate localization are also filtered out. (2) An image similarity measure function is explicitly defined and learnt to enforce the consistency between the distance of the learnt features and the underlying prostate alignment. (3) An online learning mechanism is used to adaptively integrate both the inter-patient and patient-specific information to localize the prostate. Based on the learnt image similarity measure function, the planning image of the underlying patient is aligned to the new treatment image for segmentation. The proposed method is evaluated on 163 3D prostate CT images of 10 patients, and promising experimental results are obtained.

Image-Guided Radiotherapy with Machine Learning

“…To further evaluate the performance of the proposed method, the results of several state-of-the-art methods are illustrated for comparison (see Table 9.6), which include deformable-model-based methods [13,45], registration-based methods [22,37], and learning-based methods [7,43]. From the results listed in Table 9.6, we can find that the proposed method outperforms the related methods in terms of higher mean Dice ratio and median TPF.…”

“…Note that, before segmentation on the current treatment image, the physician only needs to spend a few seconds to specify just the first and last slices of the prostate region in the CT image. With this minimal user interaction, the segmentation results can be significantly improved, compared with the fully automatic methods [7,37]. Fig.…”

“…Previous learning-based methods [7,37] first collect the voxels from certain slices and then conduct both the feature selection and the subsequent prostate-likelihood estimation for all voxels in those selected slices jointly. However, different local regions usually prefer choosing different features to better discriminate between their prostate and non-prostate voxels, as indicated by a typical example in Fig.…”

A survey of prostate modeling for image analysis

“…Liao and Shen [13] used online learning to integrate both inter-and intrapatient variations in information to localize the prostate using a sigmoid function. Contextual information was considered and is defined as 'any information that can be used to characterize the situation of an entity'.…”

“…These statistics are computed from the histogram of the gray levels of the image and depend only on the individual pixels and not on their neighborhood. In the prostate, first-order statistics have been used in the following models: the intensity profile model [47], gradient models [48,49], models using the gray level threshold of the regions extracted from a neural network [50], a radial basis relief model [51], an instantaneous variation coefficient (ICOV) model [52], a model using the local standard deviation in a multiresolution framework [53], posterior probability models [9,54], mixture probability distribution models [42,43,55,56], or models that are combined in many other ways [13,15,57,58]. Feng et al [49] proposed a weighted combination of gradient and probability distribution functions.…”