AHU-MultiNet: Adaptive loss balancing based on homoscedastic uncertainty in multi-task medical image segmentation network.

Medical image segmentation is an important field in medical image analysis and a vital part of computer-aided diagnosis. Due to the challenges in acquiring image annotations, semi-supervised learning has attracted high attention in medical image segmentation. Despite their impressive performance, most existing semi-supervised approaches lack attention to ambiguous regions (e.g., some edges or corners around the organs). To achieve better performance, we propose a novel semi-supervised method called Adaptive Loss Balancing based on Homoscedastic Uncertainty in Multi-task Medical Image Segmentation Network (AHU-MultiNet). This model contains the main task for segmentation, one auxiliary task for signed distance, and another auxiliary task for contour detection. Our multi-task approach can effectively and sufficiently extract the semantic information of medical images by auxiliary tasks. Simultaneously, we introduce an inter-task consistency to explore the underlying information of the images and regularize the predictions in the right direction. More importantly, we notice and analyze that searching an optimal weighting manually to balance each task is a difficult and time-consuming process. Therefore, we introduce an adaptive loss balancing strategy based on homoscedastic uncertainty. Experimental results show that the two auxiliary tasks explicitly enforce shape-priors on the segmentation output to further generate more accurate masks under the adaptive loss balancing strategy. On several standard benchmarks, the 2018 Atrial Segmentation Challenge and the 2017 Liver Tumor Segmentation Challenge, our proposed method achieves improvements and outperforms the new state-of-the-art in semi-supervised learning.

Adaptive Discrete Hypergraph Matching.

This paper addresses the problem of hypergraph matching using higher-order affinity information. We propose a solver that iteratively updates the solution in the discrete domain by linear assignment approximation. The proposed method is guaranteed to converge to a stationary discrete solution and avoids the annealing procedure and ad-hoc post binarization step that are required in several previous methods. Specifically, we start with a simple iterative discrete gradient assignment solver. This solver can be trapped in an -circle sequence under moderate conditions, where is the order of the graph matching problem. We then devise an adaptive relaxation mechanism to jump out this degenerating case and show that the resulting new path will converge to a fixed solution in the discrete domain. The proposed method is tested on both synthetic and real-world benchmarks. The experimental results corroborate the efficacy of our method.

Tongue Images Classification Based on Constrained High Dispersal Network.

Computer aided tongue diagnosis has a great potential to play important roles in traditional Chinese medicine (TCM). However, the majority of the existing tongue image analyses and classification methods are based on the low-level features, which may not provide a holistic view of the tongue. Inspired by deep convolutional neural network (CNN), we propose a novel feature extraction framework called constrained high dispersal neural networks (CHDNet) to extract unbiased features and reduce human labor for tongue diagnosis in TCM. Previous CNN models have mostly focused on learning convolutional filters and adapting weights between them, but these models have two major issues: redundancy and insufficient capability in handling unbalanced sample distribution. We introduce high dispersal and local response normalization operation to address the issue of redundancy. We also add multiscale feature analysis to avoid the problem of sensitivity to deformation. Our proposed CHDNet learns high-level features and provides more classification information during training time, which may result in higher accuracy when predicting testing samples. We tested the proposed method on a set of 267 gastritis patients and a control group of 48 healthy volunteers. Test results show that CHDNet is a promising method in tongue image classification for the TCM study.

Effective identification and localization of immature precursors in bone marrow biopsy.

Abnormal localization of immature precursors (ALIP) aggregating and clustering in bone marrow biopsy appears earlier than that of bone marrow smears in detection of the relapse of acute myelocytic leukemia (AML). But traditional manual ALIP recognition has many shortcomings such as prone to false alarms, neglect of distribution law before three immature precursor cells gathered, and qualitative analysis instead of quantitative one. So, it is very important to develop a novel automatic method to identify and localize immature precursor cells for computer-aided diagnosis, to disclose their patterns before ALIP with the development of AML. The contributions of this paper are as follows. (1) After preprocessing the image with Otsu method, we identify both precursor cells and trabecular bone by multiple morphological operations and thresholds. (2) We localize the precursors in different regions according to their distances with the nearest trabecular bone based on chamfer distance transform, followed by discussion for the presumptions and limitations of our method. The accuracy of recognition and localization is evaluated based on a comparison with visual evaluation by two blinded observers.

A random forest model based classification scheme for neonatal amplitude-integrated EEG.

BACKGROUND: Modern medical advances have greatly increased the survival rate of infants, while they remain in the higher risk group for neurological problems later in life. For the infants with encephalopathy or seizures, identification of the extent of brain injury is clinically challenging. Continuous amplitude-integrated electroencephalography (aEEG) monitoring offers a possibility to directly monitor the brain functional state of the newborns over hours, and has seen an increasing application in neonatal intensive care units (NICUs). METHODS: This paper presents a novel combined feature set of aEEG and applies random forest (RF) method to classify aEEG tracings. To that end, a series of experiments were conducted on 282 aEEG tracing cases (209 normal and 73 abnormal ones). Basic features, statistic features and segmentation features were extracted from both the tracing as a whole and the segmented recordings, and then form a combined feature set. All the features were sent to a classifier afterwards. The significance of feature, the data segmentation, the optimization of RF parameters, and the problem of imbalanced datasets were examined through experiments. Experiments were also done to evaluate the performance of RF on aEEG signal classifying, compared with several other widely used classifiers including SVM-Linear, SVM-RBF, ANN, Decision Tree (DT), Logistic Regression(LR), ML, and LDA. RESULTS: The combined feature set can better characterize aEEG signals, compared with basic features, statistic features and segmentation features respectively. With the combined feature set, the proposed RF-based aEEG classification system achieved a correct rate of 92.52% and a high F1-score of 95.26%. Among all of the seven classifiers examined in our work, the RF method got the highest correct rate, sensitivity, specificity, and F1-score, which means that RF outperforms all of the other classifiers considered here. The results show that the proposed RF-based aEEG classification system with the combined feature set is efficient and helpful to better detect the brain disorders in newborns.

[Characteristics of "pre-ALIP" in bone marrow sections of patients with acute myeloid leukemia].

To detect the characteristics of "pre-ALIP" and to investigate their relevance with the development of acute myeloid leukemia (AML) by computer image procession technology, bone marrow (BM) was collected by aspiration/trephine biopsy from AML patients during the complete remission (CR). BM sections were stained by HGF (haematoxylin-Giemsa-acid fuchsin) and photographed by optical microscope imaging system. 4 kinds of computer image segmentation technologies were compared to select the best one for detecting the localization and quantitation of the precursor cells. Planimetry was combined with morphology to segment bone trabeculae. The number of single and double-cluster precursor cells and their distance from bone trabeculae was detected with Euclidean distance change method in BM images of AML patients, and compared with the normal controls. Moreover, the morphological characteristics of "pre-ALIP" were investigated, and the correlation with the development of AML was analyzed. The results showed that the computer image segmentation method based on morphology could identify the precursor cells and bone trabeculae more exactly in BM image, as compared with the methods of 8-Sobel operater. Canny operator and watershed algorithm. Bone trabeculae could be segmented with combinative methods of morphology and planimetry. The number of single precursor cells (19.27 ± 11.60)/mm(2) and double-cluster precursor cells (1.77 ± 1.76)/mm(2) in CR group were higher than that in normal controls (p < 0.05). The distance of single precursor cells from bone trabeculae in CR group were closer to bone trabeculae than that in controls [(230.12 ± 97.68) µm vs (260.92 ± 99.88 µm)] (p < 0.05), but the distance of double-cluster precursor cells from bone trabeculae in AML patients was (274.56 ± 139.48) µm, which showed no statistically significant different from controls (p > 0.05), while the double-cluster precursor cells showed the tendency of migrating to the intermediate zone of bone trabeculae compared with the single precursor cells in CR group (p < 0.05). It is concluded that the structure of "pre-ALIP" in BM tissue exists before the occurrence of ALIP. The characteristics of "pre-ALIP" are single and double-cluster precursor cells with abnormal localization or quantitation, which showed correlation with the development of AML.

Ultrasonic liver discrimination using 2-D phase congruency.

In this paper, we present an experiment to extract liver features using two-dimensional phase congruency, which is invariant to changes in intensity or contrast, to try to avoid the influence of machine settings. The effectiveness of our method was tested on three classes of liver images and shows the potential for physicians to quantify liver pathology in clinical diagnosis.

Liver fibrosis identification based on ultrasound images captured under varied imaging protocols.

Diagnostic ultrasound is a useful and noninvasive method in clinical medicine. Although due to its qualitative, subjective and experience-based nature, ultrasound image interpretation can be influenced by image conditions such as scanning frequency and machine settings. In this paper, a novel method is proposed to extract the liver features using the joint features of fractal dimension and the entropies of texture edge co-occurrence matrix based on ultrasound images, which is not sensitive to changes in emission frequency and gain. Then, Fisher linear classifier and support vector machine are employed to test a group of 99 in-vivo liver fibrosis images from 18 patients, as well as other 273 liver images from 18 normal human volunteers.

Liver fibrosis identification based on ultrasound images.

Diagnostic ultrasound is one of useful and noninvasive tools for clinical medicine. However, due to its qualitative, subjective and experience-based nature, ultrasound images can be influenced by image conditions such as scanning frequency and machine settings. In this paper, a novel method is proposed to extract the liver features using the joint features of fractal dimension and the entropies of texture edge co-occurrence matrix based on ultrasound images, which is not sensitive to changes in emission frequency and gain. Then, Fisher linear classifier and Support Vector Machine are employed to test on a group of 99 liver fibrosis images from 18 patients, as well as other 273 healthy liver images from 18 specimens.

Ultrasonic liver characterization using phase congruency.

Feature extraction plays an important role in the whole process of liver characterization. Because the ultrasonic scanner in use can be adjusted by different clinicians to produce optimal images, the ultrasound images captured sometimes can be greatly influenced by machine settings and further impact the classification result. In this paper, some experiments are made to try to extract the liver features using the 2D phase congruency, which invariant to changes in intensity or contrast, to try to avoid those problems. The effectiveness of our method tested on three classes of liver images shows the potential for physicians to quantify liver status in clinical diagnosis.