Average Shape Model Research Articles

Prior-based 3D U-Net: A model for knee-cartilage segmentation in MRI images

In knee-joint magnetic resonance imaging (MRI), articular cartilage is susceptible to artifacts and noise, making it difficult for a single image segmentation algorithm to achieve accurate segmentation. To overcome this challenge, we propose a 3D U-Net neural network method based on prior knowledge (Prior-based 3D U-Net). First, sample data are used to train the 3D U-Net model and calculate an average shape model (ASM) of the knee joint (including the cartilage, distal femur, and proximal tibia). Second, 3D U-Net is used to segment the distal femur and proximal tibia from newly input knee MRI data. Third, we register the newly input knee joints (the distal femur and proximal tibia) with the ASM by constructing a registration transformation RT. Fourth, RT converts ASM into the newly input MRI images, and a predicted cartilage model appears. Finally, the predicted cartilage model, distal femur, and proximal tibia segmented by the previous U-Net step are used as constraints of 3D U-Net to segment the knee cartilage. According to the cartilage segmentation results, the Dice coefficient (DSC) of Prior-based 3D U-Net algorithm was 74.02%, the intersection-over-union (IoU) coefficient was 57.56%, the average symmetric surface distance (ASD) distance was 0.51 mm, the 95% Hausdorff distance (Hd95) was 3.89 mm, and the segmentation time for one MRI image was 0.38 s on average. These quantitative indicators are higher than those obtained from similar U-Net models and fully convolutional networks (FCN) models, indicating that our algorithm could achieve more accurate automatic cartilage segmentation, greatly reducing medical researchers’ burden.

Online Hard Patch Mining Using Shape Models and Bandit Algorithm for Multi-Organ Segmentation.

Hard sample selection can effectively improve model convergence by extracting the most representative samples from a training set. However, due to the large capacity of medical images, existing sampling strategies suffer from insufficient exploitation for hard samples or high time cost for sample selection when adopted by 3D patch-based models in the field of multi-organ segmentation. In this paper, we present a novel and effective online hard patch mining (OHPM) algorithm. In our method, an average shape model that can be mapped with all training images is constructed to guide the exploration of hard patches and aggregate feedback from predicted patches. The process of hard mining is formalized as a multi-armed bandit problem and solved with bandit algorithms. With the shape model, OHPM requires negligible time consumption and can intuitively locate difficult anatomical areas during training. The employment of bandit algorithms ensures online and sufficient hard mining. We integrate OHPM with advanced segmentation networks and evaluate them on two datasets containing different anatomical structures. Comparative experiments with other sampling strategies demonstrate the superiority of OHPM in boosting segmentation performance and improving model convergence. The results in each dataset with each network suggest that OHPM significantly outperforms other sampling strategies by nearly 2% average Dice score.

Group-wise Shape Correspondence of Variable and Complex Objects.

We present a group-wise shape correspondence method for analyzing variable and complex objects in a population study. The proposed method begins with the standard spherical harmonics (SPHARM) point distribution models (PDM) with their spherical mappings. In case of complex and variable objects, the equal area spherical mapping based SPHARM correspondence is imperfect. For such objects, we present here a novel group-wise correspondence. As an example dataset, we use 12 second mandibular molars representing 6 living or fossil euarchontan species. To improve initial correspondence of the SPHARM-PDM representation, we first apply a rigid transformation on each subject using five well-known landmarks (molar cusps). We further enhance the correspondence by optimizing landmarks (local) and multidimensional geometric property (global) over each subject with spherical harmonic representation. The resulting average shape model better captures sharp landmark representation in quantitative evaluation as well as a nice separation of different species compared with that of the SPHARM-PDM method.

초음파 전립선 영상에서 전립선 경계 분할을 위한 평균 형상 모델

전립선암은 전립선에 나타나는 악성 종양이다. 현재 그 발병률이 높아지고 있다. 전립선암의 구조를 가장 정확하게 확인할 수 있는 검사 방법은 MRI를 이용하는 것이나, 그 비용 때문에 모든 환자에게 적용하기는 어려운 실정이다. 그래서 많은 환자들은 가격이 저렴한 초음파검사를 이용하여 전립선암을 진단하고 있다. 전통적으로 의사들은 영상을 눈으로 확인하여 전립선의 경계를 수동으로 분할하였다. 그러나 수동으로 분할하는 과정은 시간이 많이 소요되며, 의사에 따라서 그 경계가 일정하지 않게 얻어진다. 이 문제를 해결하기 위하여 전립선의 자동 분할에 관한 연구가 되었고, 환자들에게 신뢰를 줄 수 있었다. 본 연구는 초음파 전립선 영상에서 전립선의 경계를 분할하는데 평균 형상 모델을 적용하는 것이다. 먼저, 에지 분포를 이용하여 프로브를 찾고, 프로브와 연결된 두 직선을 찾는다. 이 후에 이 정보를 이용하여 전립선 영상 위에 평균 형상을 위치시킨다. Prostate cancer is a malignant tumor occurring in the prostate. Recently, the repetition rate is increasing. Image inspection method which we can check the prostate structure the most correctly is MRI(Magnetic Resonance Imaging), but it is hard to apply it to all the patients because of the cost. So, they use mostly TRUS(Transrectal Ultrasound) images acquired from prostate ultrasound inspection and which are cheap and easy to inspect the prostate in the process of treating and diagnosing the prostate cancer. Traditionally, in the hospital the doctors saw the TRUS images by their eyes and manually segmented the boundary between the prostate and nonprostate. But the manually segmenting process not only needed too much time but also had different boundaries according to the doctor. To cope the problems, some automatic segmentations of the prostate have been studied to generate the constant segmentation results and get the belief from patients. In this study, we propose an average shape model to segment the prostate boundary in TRUS prostate image. The method has 3 steps. First, it finds the probe using edge distribution. Next, it finds two straight lines connected with the probe. Finally it puts the shape model to the image using the position of the probe and straight lines.

평균 형상 모델과 SIFT 특징을 이용한 TRUS 영상의 전립선 분할

전립선암은 남자에게 가장 흔히 나타나는 암 중의 하나이며, 많은 나라에서 죽음에 이르게 하는 큰 요인이 되고 있다. 전립선암을 진단하고 치료하는 과정에서 비용이 싼 TRUS 영상이 사용된다. 그러나 전립선 경계의 정확한 구분이 요구되지만 어려운 문제이다. 그 이유는 경계가 불명확하고, 반점들이 많으며, 그레이 레벨의 범위가 작기 때문이다. 본 연구에서는 전립선의 평균 형상 모델과 불변의 특징을 이용하여 TRUS 영상에서 자동으로 전립선 분할하는 방법을 제안한다. 이 방법은 4 단계로 구성된다. 먼저, 에지 분포를 이용하여 프로브와 두개의 직선을 찾아낸다. 다음으로, 평균 형상 모델의 중앙에 위치한 3개의 전립선 패치를 획득한다. 이 패치는 전립선과 비전립선의 특징을 비교하기 위해 사용된다. 다음으로, 세 개의 패치와 각 블록들이 얼마나 대표 블록과 유사한지를 비교한다. 마지막으로, 앞 단계의 경계와 첫 단계에서 얻은 개략적 경계가 최종 분할에 사용된다. 이 방법의 유효성을 검증하기 위하여 실험을 하였으며, 인간 전문가에 의해 얻어진 경계와 비교하여 7.78% 미만의 차이로 경계를 얻을 수 있었다. Prostate cancer is one of the most frequent cancers in men and is a major cause of mortality in the most of countries. In many diagnostic and treatment procedures for prostate disease, transrectal ultrasound(TRUS) images are being used because the cost is low. But, accurate detection of prostate boundaries is a challenging and difficult task due to weak prostate boundaries, speckle noises and the short range of gray levels. This paper proposes a method for automatic prostate segmentation in TRUS images using its average shape model and invariant features. This approach consists of 4 steps. First, it detects the probe position and the two straight lines connected to the probe using edge distribution. Next, it acquires 3 prostate patches which are in the middle of average model. The patches will be used to compare the features of prostate and nonprostate. Next, it compares and classifies which blocks are similar to 3 representative patches. Last, the boundaries from prior classification and the rough boundaries from first step are used to determine the segmentation. A number of experiments are conducted to validate this method and results showed that this new approach extracted the prostate boundary with less than 7.78% relative to boundary provided manually by experts.