Domain Adaptation for Medical Image Segmentation: A Meta-Learning Method.

Convolutional neural networks (CNNs) have demonstrated great achievement in increasing the accuracy and stability of medical image segmentation. However, existing CNNs are limited by the problem of dependency on the availability of training data owing to high manual annotation costs and privacy issues. To counter this limitation, domain adaptation (DA) and few-shot learning have been extensively studied. Inspired by these two categories of approaches, we propose an optimization-based meta-learning method for segmentation tasks. Even though existing meta-learning methods use prior knowledge to choose parameters that generalize well from few examples, these methods limit the diversity of the task distribution that they can learn from in medical image segmentation. In this paper, we propose a meta-learning algorithm to augment the existing algorithms with the capability to learn from diverse segmentation tasks across the entire task distribution. Specifically, our algorithm aims to learn from the diversity of image features which characterize a specific tissue type while showing diverse signal intensities. To demonstrate the effectiveness of the proposed algorithm, we conducted experiments using a diverse set of segmentation tasks from the Medical Segmentation Decathlon and two meta-learning benchmarks: model-agnostic meta-learning (MAML) and Reptile. U-Net and Dice similarity coefficient (DSC) were selected as the baseline model and the main performance metric, respectively. The experimental results show that our algorithm maximally surpasses MAML and Reptile by 2% and 2.4% respectively, in terms of the DSC. By showing a consistent improvement in subjective measures, we can also infer that our algorithm can produce a better generalization of a target task that has few examples.

Advanced Warning of Aortic Dissection on Non-Contrast CT: The Combination of Deep Learning and Morphological Characteristics.

Background: The identification of aortic dissection (AD) at baseline plays a crucial role in clinical practice. Non-contrast CT scans are widely available, convenient, and easy to perform. However, the detection of AD on non-contrast CT scans by radiologists currently lacks sensitivity and is suboptimal. Methods: A total of 452 patients who underwent aortic CT angiography (CTA) were enrolled retrospectively from two medical centers in China to form the internal cohort (341 patients, 139 patients with AD, 202 patients with non-AD) and the external testing cohort (111 patients, 46 patients with AD, 65 patients with non-AD). The internal cohort was divided into the training cohort (n = 238), validation cohort (n = 35), and internal testing cohort (n = 68). Morphological characteristics were extracted from the aortic segmentation. A deep-integrated model based on the Gaussian Naive Bayes algorithm was built to differentiate AD from non-AD, using the combination of the three-dimensional (3D) deep-learning model score and morphological characteristics. The areas under the receiver operating characteristic curve (AUCs), accuracy, sensitivity, and specificity were used to evaluate the model performance. The proposed model was also compared with the subjective assessment of radiologists. Results: After the combination of all the morphological characteristics, our proposed deep-integrated model significantly outperformed the 3D deep-learning model (AUC: 0.948 vs. 0.803 in the internal testing cohort and 0.969 vs. 0.814 in the external testing cohort, both p < 0.05). The accuracy, sensitivity, and specificity of our model reached 0.897, 0.862, and 0.923 in the internal testing cohort and 0.730, 0.978, and 0.554 in the external testing cohort, respectively. The accuracy for AD detection showed no significant difference between our model and the radiologists (p > 0.05). Conclusion: The proposed model presented good performance for AD detection on non-contrast CT scans; thus, early diagnosis and prompt treatment would be available.

Potential value of the PixelShine deep learning algorithm for increasing quality of 70 kVp+ASiR-V reconstruction pelvic arterial phase CT images.

OBJECTIVE: To investigate the effect of a deep learning-based denoising algorithm, PixelShine (PS), on the quality of 70 kVp pelvic arterial phase CT images. MATERIALS AND METHODS: A retrospective analysis was performed on arterial phase pelvic CT images from 33 patients (body-mass index ≤ 20 kg/m2) obtained with a GE Revolution CT (70 kVp tube voltage; adaptive statistical iterative reconstruction-Veo-filtered back projection, 50% blending) and designated group A. Group B images were then obtained by applying PS to group A image datasets. Subjective image quality was evaluated by two radiologists with a 5-point scoring system; the scores of the groups were compared. Image signal was assessed using CT values of the urinary bladder. CT and standard deviation (SD) values of the gluteus maximus were measured, and SD values of the gluteus maximus were used to represent image noise. The signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of the bladder were calculated. Image noise, SNR, and CNR of two groups were compared using paired t-tests. RESULTS: The subjective visual image quality scores of groups A and B, respectively, were 3.11 ± 0.30 vs. 3.82 ± 0.57; image noise was 15.79 ± 2.05 Hounsfield units (HU) vs. 11.06 ± 2.22 HU; SNRs of bladder were 0.50 ± 0.23 vs. 0.79 ± 0.39; and CNRs of bladder were 3.72 ± 0.85 vs. 5.14 ± 1.27. Group B showed better subjective image quality, lower image noise, and improved SNR and CNR, compared to group A; these differences were statistically significant (P < 0.05). The noise of group B was approximately 30% lower than that of group A; the SNR and CNR values of group B were improved by approximately 58% and 38%, respectively. CONCLUSION: Using 70 kVp +ASiR-V, PS can improve the image quality of pelvic arterial phase CT images, significantly reduce the image noise, and improve the SNR and CNR.

Imaging lesions of the lateral hip.

The lateral aspect of the hip is composed of a complex array of osseous and soft tissue structures. Both common and uncommon clinical entities are encountered in the lateral hip. This article briefly introduces fundamental imaging anatomy and the functional roles of different osseous and soft tissue structures in the lateral aspect of the hip, followed by a discussion of relevant imaging findings of lateral hip pathology. Greater trochanteric pain syndrome is frequently encountered in patients with lateral hip pain and encompasses a spectrum of soft tissue abnormalities including trochanteric and subgluteal bursitis, and tendinopathy or tears of the gluteal tendons. In addition, different types of injuries to the gluteal myotendinous unit and injuries to the indirect head of the rectus femoris, proximal iliotibial band, and the lateral joint capsular ligaments can present with lateral hip pain. Some of the less common soft tissue abnormalities of the lateral hip include Morel-Lavallée lesion and meralgia paresthetica.

Imaging of soft tissue abnormalities about the hip.

Soft tissue abnormalities about the hip represent a common clinical problem. Although the signs and symptoms of some of these abnormalities are clinically evident, other entities are frequently overlooked. This article provides an overview and discusses the role of major imaging modalities, especially MR imaging, the primary modality for evaluation of many soft-tissue abnormalities. An introduction to fundamental imaging anatomy and functional roles of soft tissue structures about the hip is provided, recognizing their importance in making the correct diagnosis. Intra-articular and extra-articular soft tissue abnormalities reviewed systematically according to their mechanism of injury and anatomic or functional compartments.

Knee cartilage T2 characteristics and evolution in relation to morphologic abnormalities detected at 3-T MR imaging: a longitudinal study of the normal control cohort from the Osteoarthritis Initiative.

PURPOSE: To determine the frequency of degenerative knee morphologic abnormalities in asymptomatic individuals by using 3-T magnetic resonance (MR) imaging and to investigate the characteristics and evolution of cartilage T2 values in relation to morphologic abnormalities with a longitudinal study. MATERIALS AND METHODS: The study was approved by the institutional review board and was compliant with HIPAA. Ninety-five asymptomatic subjects aged 45-78 years who were free of risk factors for osteoarthritis (OA) were selected from the Osteoarthritis Initiative normal control cohort and examined with radiography and 3-T MR imaging. Data obtained at both baseline and 2-year follow-up were analyzed. OA-related knee abnormalities were analyzed by using the whole-organ MR imaging score (WORMS). Cartilage T2 maps were generated by using sagittal two-dimensional multiecho spin-echo images of the right knee. Statistical significance was determined with the Student t test, the paired t test, a mixed random effects model, one-way analysis of variance, and a multiple linear regression model. RESULTS: Knee abnormalities were identified with a high frequency (90% at baseline and 92% at 2-year follow-up). The prevalence of hyaline cartilage lesions was particularly high (86% at baseline and 84% at follow-up). A significant longitudinal increase in T2 was detected in the tibiofemoral cartilage but not the patellofemoral cartilage (P = .0072). The longitudinal change in T2 was significantly associated with worsening of the cartilage WORMS (P = .038). CONCLUSION: Asymptomatic subjects have a high frequency of OA-related morphologic abnormalities. A significant increase in tibiofemoral cartilage T2 was detected over the 2-year period. A greater increase in T2 was associated with increased progression of cartilage morphologic abnormalities.

Meniscal T1rho and T2 measured with 3.0T MRI increases directly after running a marathon.

PURPOSE: To prospectively evaluate changes in T1rho and T2 relaxation time in the meniscus using 3.0 T MRI in asymptomatic knees of marathon runners and to compare these findings with those of age-matched healthy subjects. MATERIAL AND METHODS: Thirteen marathon runners underwent 3.0 T MRI including T1rho and T2 mapping sequences before, 48-72 h after, and 3 months after competition. Ten controls were examined at baseline and after 3 months. All images were analyzed by two musculoskeletal radiologists identifying and grading cartilage, meniscal, ligamentous. and other knee abnormalities with WORMS scores. Meniscal segmentation was performed to generate T1rho and T2 maps in six compartments. RESULTS: No differences in morphological knee abnormalities were found before and after the marathon. However, all marathon runners showed a significant increase in T1rho and T2 values after competition in all meniscus compartments (p < 0.0001), which may indicate changes in the biochemical composition of meniscal tissue. While T2 values decreased after 3 months T1rho values remained at a high level, indicating persisting changes in the meniscal matrix composition after a marathon. CONCLUSION: T2 values in menisci have the potential to be used as biomarkers for identifying reversible meniscus matrix changes indicating potential tissue damage. T1rho values need further study, but may be a valuable marker for diagnosing early, degenerative changes in the menisci following exercise.

Identification of a novel common proviral integration site, flit-1, in feline leukemia virus induced thymic lymphoma.

A new proviral integration site for feline leukemia virus (FeLV), termed flit-1, was identified from feline thymic lymphoma. Among 35 FeLV-related tumors examined, 5 of 25 thymic lymphomas demonstrated proviral insertion within flit-1 locus whereas none of four alimentary and five multicentric lymphomas and one T-lymphoid leukemia examined had rearrangement in this region. Extensive sequence analysis has shown that flit-1, which is noncoding, is conserved on human chromosome 12 and mouse chromosome 15. The human and murine homologs of flit-1 are positioned approximately 30-kb upstream to activin-A receptor type II-like 1 (ACVRL1/ALK1) gene. Expression of ACVRL1 mRNA was examined in two of five lymphomas with flit-1 rearrangement and detected in both of the two whereas normal thymuses and seven lymphoid tumors without flit-1 rearrangement had no detectable expression. Therefore, flit-1 appears to represent a novel FeLV proviral common integration domain that may influence lymphomagenesis as insertional mutagenesis.

Stromal derived factor-1 (SDF-1/CXCL12) and CXCR4 in renal cell carcinoma metastasis.

Renal cell carcinoma (RCC) is characterized by organ-specific metastases. The chemokine stromal derived factor-1 (SDF-1/CXCL12) and its receptor CXCR4 have been suggested to regulate organ-specific metastasis in various other cancers. On this basis, we hypothesized that the biological axis of CXCL12 via interaction with its receptor, CXCR4, is a major mechanism for RCC metastasis. We demonstrated that CXCR4 was significantly expressed on circulating cytokeratin+ RCC cells from patients with known metastatic RCC. We detected up-regulation of CXCR4 mRNA and protein levels on a human RCC cell line by either knockdown of the von Hippel-Lindau (VHL) tumor suppressor protein, or incubating the cells under hypoxic conditions. The enhanced CXCR4 expression was mediated through the interaction of the Hypoxia Inducible Factor-1alpha (HIF-1alpha) with the promoter region of the CXCR4 gene. Furthermore, the expression of CXCR4 on human RCC directly correlated with their metastatic ability in vivo in both heterotopic and orthotopic SCID mouse models of human RCC. Neutralization of CXCL12 in SCID mice abrogated metastasis of RCC to target organs expressing high levels of CXCL12; without altering tumor cell proliferation, apoptosis, or tumor-associated angiogenesis. Therefore, our data suggest that the CXCL12/CXCR4 biological axis plays an important role in regulating the organ-specific metastasis of RCC.

CXCR3/CXCR3 ligand biological axis impairs RENCA tumor growth by a mechanism of immunoangiostasis.

Metastatic renal cell carcinoma (RCC) responds poorly to chemo- or radiation therapy but appears to respond to systemic immunotherapy (i.e., IL-2 and/or IFN-alpha), albeit with only 5-10% durable response. The CXCR3/CXCR3 ligand biological axis plays an important role in mediating type 1 cytokine-dependent cell-mediated immunity, which could be beneficial for attenuating RCC if optimized. We found that systemic IL-2 induced the expression of CXCR3 on circulating mononuclear cells but impaired the CXCR3 ligand chemotactic gradient from plasma to tumor by increasing circulating CXCR3 ligand levels in a murine model of RCC. Moreover, the antitumor effect of systemic IL-2 was CXCR3-dependent, as IL-2 failed to inhibit tumor growth and angiogenesis in CXCR3-/- mice. We hypothesized that the immunotherapeutic effect of the CXCR3/CXCR3 ligand biological axis could be optimized by first priming with systemic IL-2 to induce CXCR3 expression on circulating mononuclear cells followed by enhancing the intratumor CXCR3 ligand levels to establish optimal CXCR3-dependent chemotactic gradient. We found that combined systemic IL-2 with an intratumor CXCR3 ligand (CXCL9) lead to significantly greater reduction in tumor growth and angiogenesis, increased tumor necrosis, and increased intratumor infiltration of CXCR3+ mononuclear cells, as compared with either IL-2 or CXCL9 alone. The enhanced antitumor effect of the combined strategy was associated with a more optimized CXCR3-dependent chemotactic gradient and increased tumor-specific immune response. These data suggest that the combined strategy of systemic IL-2 with intratumor CXCR3 ligand is more efficacious than either strategy alone for reducing tumor-associated angiogenesis and augmenting tumor-associated immunity, the concept of immunoangiostasis.

In vivo expression of GFP transgene delivered via a replicating feline leukemia virus.

We previously described replication-competent feline leukemia virus (FeLV) vectors with high-level and stable expression of a green fluorescent protein (GFP) or a suicide transgene in cell cultures in vitro. Considering that FeLV might potentially be used to deliver therapeutic genes in vivo, we first evaluated the expression of the GFP gene introduced in cats by the FeLV, Rickard subgroup A (FRA) construct. Eight newborn kittens were either inoculated with pFRA-GFP plasmid DNA intradermally, or challenged intraperitoneally with FRA-GFP-infected feline fibroblasts. During a 12-week observation period, five cats were shown to be progressively viremic. Quantitative PCR and RT-PCR analyses of plasma and tissue samples from these cats showed that GFP was retained in FeLV DNA or RNA to a variable degree, ranging from 0.002 to 27.890%. Tissue DNA samples were analyzed by PCR for the status of GFP and the env-transgene complex. While the proviruses carrying the GFP transgene were shown to be minor species, all tissues, however, retained the full-length GFP transgene. Despite the occurrence of predominant species with various deletions in the viral genome, approximately 1-3% of the total cell population was GFP-positive in the lymphoid tissues as visualized by laser confocal microscopy. Co-localization of immunofluorescent cells indicated that CD3-positive T cells, dendritic cells and macrophages were the major targets for GFP expression. These findings on the detectable in vivo expression of GFP for as long as a period of 3 months could be viewed positively for contemplating a therapeutic strategy for control of FeLV infection in the cats.

A potential therapeutic strategy to combat leukemia virus infection.

To test the concept that a replication-competent retrovirus carrying a suicide gene could have potential utility in the control of the natural virus infection in mammalian species, we constructed derivatives of a feline leukemia virus (FeLV) that is commonly associated with leukemia-lymphomas in this species. The FeLV, Rickard strain, subgroup A (FRA) genome contained at the 3' end of the envgene, an insert of an internal ribosomal entry site (IRES) linked to cDNA sequence of either herpes simplex virus thymidine kinase (HSV-TK) or a truncated HSV-TK (HSV-ATK) or yeast cytosine deaminase (CD). These constructs were transfected into feline fibroblast cells (H927). The viruses produced were determined to be replication-competent. The stable propagation of the full-length transgene was, however, dependent on the size of the insert, IRES-CD being the smallest in size (1031 bp) exhibiting maximal stability for at least up to six months. The protein products of the transgenes could be detected, despite the appearance of deleted proviruses at late passages. The transduced cells were susceptible to cytotoxic killing when the appropriate prodrug, ganciclovir (GCV), acyclovir (ACV) or 5-fluorocytosine (5-FC) was added to the culture medium. H927 cells, infected with another subgroup of FeLV, namely, FeLV-B or FeLV-C, could be superinfected by the FRA-suicide gene viruses and thus, subjected to killing. Interestingly, at an early stage of infection by the parental FRA, H927 cells could also be reinfected by the same subgroup FRA constructs to induce the suicide effect. Among the three constructs, the vector with the CD gene was determined to be superior to others in terms of stability, therapeutic index and bystander effect in the cell culture test system. While the in vivo correlates of the therapeutic effect in the feline model remain to be determined, our results do encourage investigation of the same concept in the control of HTLV and, perhaps even, HIV infection in humans.