GoogLeNet could judge the types of head MRI sequences with a small amount of training data, irrespective of morphological or contrast conditions.
Increased use of deep convolutional neural networks (DCNNs) in medical imaging diagnosis requires determinate evaluation of diagnostic performance. We performed the fundamental investigation of diagnostic performance of DCNNs using the detection task of brain metastasis. Methods: We retrospectively investigated AlexNet and GoogLeNet using 3117 positive and 37961 negative MRI images with and without metastasis regarding (1) diagnostic biases, (2) the optimal K number of K-fold cross validations (K-CVs), (3) the optimal positive versus negative image ratio, (4) the accuracy improvement curves, (5) the accuracy range prediction by the bootstrap method, and (6) metastatic lesion detection by regions with CNNs (R-CNNs). Results: Respectively, AlexNet and GoogLeNet had (1) 50 ± 4.6% and 50 ± 4.9% of the maximal mean ± 95% confidence intervals (95% CIs) measured with equal-sized negative versus negative image datasets and positive versus positive image datasets, (2) no less than 10 and 4 of K number in K-CVs fell within the respective maximum biases of 4.6% or 4.9%, (3) 74% of the highest accuracy with equal positive versus negative image ratio dataset and 91% of that with four times of negative-to-positive image ratio dataset, (4) the accuracy improvement curves increasing from 69% to 74% and 73% to 88% as positive versus negative pairs of the training images increased from 500 to 2495, (5) at least nine and six out of 10-CV result sets essential to predict the accuracy ranges by the bootstrap method, and (6) 50% and 45% of metastatic lesion detection accuracies by R-CNNs. Conclusions: Our research presented methodological fundamentals to evaluate diagnostic features in the visual recognition of DCNNs. Our series will help to conduct the accuracy investigation of computer diagnosis in medical imaging.
Accurate Computer-Assisted Diagnosis, associated with proper data wrangling, can alleviate the risk of overlooking the diagnosis in a clinical environment. Towards this, as a Data Augmentation (DA) technique, Generative Adversarial Networks (GANs) can synthesize additional training data to handle the small/fragmented medical imaging datasets collected from various scanners; those images are realistic but completely different from the original ones, filling the data lack in the real image distribution. However, we cannot easily use them to locate disease areas, considering expert physicians' expensive annotation cost. Therefore, this paper proposes Conditional Progressive Growing of GANs (CPGGANs), incorporating highly-rough bounding box conditions incrementally into PGGANs to place brain metastases at desired positions/sizes on 256 × 256 Magnetic Resonance (MR) images, for Convolutional Neural Network-based tumor detection; this first GAN-based medical DA using automatic bounding box annotation improves the training robustness. The results show that CPGGAN-based DA can boost 10% sensitivity in diagnosis with clinically acceptable additional False Positives. Surprisingly, further tumor realism, achieved with additional normal brain MR images for CPGGAN training, does not contribute to detection performance, while even three physicians cannot accurately distinguish them from the real ones in Visual Turing Test.
Background and Purpose: During a percutaneous vertebroplasty (PVP) procedure, patients typically lie in the prone position. However, some elderly patients have difficulty in maintaining the prone position. Therefore, we aimed to investigate the safety and efficacy of PVP in a lateral decubitus position in patients experiencing difficulty in maintaining the prone position. Materials and Methods: A total of 123 PVP procedures performed consecutively on 117 patients for symptomatic vertebral fractures caused by bone tumors or osteoporosis were studied. The patients were divided into prone (n=113) and decubitus groups (n=10) according to their positions during the PVP procedures. The factors related to the patients' background, procedures, therapeutic effects, and adverse events were compared between the 2 groups. Univariate analysis was performed using Student's t-test, Mann-Whitney's U-test, chisquared test, or Wilcoxon signed-rank test. Results: In the decubitus group, the average age was 6.7 years older (p<0.05), the average setup time was 1.6 times longer (p<0.01), the average fluoroscopic exposure dose was 1.37 times greater (p<0.05), the average dose-length product of interventional computed tomography was 1.78 times greater (p<0.05), and mobility restoration on the 7 th day after the PVP was less (p<0.05) compared to the prone group. There were no significant differences in bone cement leakage, pulmonary embolism, recurrence of compression fractures, or pain relief. Conclusion: Although some disadvantages were observed, decubitus PVP seemed to be completed safely and successfully. Decubitus PVP can become a treatment option for patients with vertebral fractures and difficulty in maintaining the prone position.
In hospitalized coronavirus disease 2019 (COVID-19) patients, anticoagulation therapy is administered to prevent thrombosis. However, anticoagulation sometimes causes bleeding complications. We herein report two Japanese cases of severe COVID-19 in which spontaneous muscle hematomas (SMH) developed under therapeutic anticoagulation with unfractionated heparin. Although the activated partial prothrombin time was within the optimal range, contrast-enhanced computed tomography (CECT) revealed SMH in the bilateral iliopsoas muscles in both cases, which required emergent transcatheter embolization. Close monitoring of the coagulation system and the early diagnosis of bleeding complications through CECT are needed in severe COVID-19 patients treated with anticoagulants.
Purpose Using the multi-detector computed tomography and related three-dimensional imaging technology, we developed a vertebral needle targeting simulation training system named spinal needling intervention practice using ray-summation imaging (SNIPURS). Herein, we assessed the utility of SNIPURS by evaluating changes in the learning curves of SNIPURS trainees. Methods Twenty-one examinees were enrolled: seven experienced operators (expert group), seven trainees with coaching (coaching group), and seven trainees without coaching (non-coaching group). They performed six tests of vertebral needle targeting simulation on the workstation-generated spinal ray-summation images of six patients with vertebral fractures. In each test, they determined the bilateral trans-pedicular puncture points and angles on two thoracic and two lumbar vertebrae on ray-summation imaging (i.e., 8 simulations per test). The coaching group received coaching by a trainer after Tests 1 and 4, while the others did not. Scores were given based on the trans-pedicular pathway (1 point) or not (0 point). Eight virtual needles were evaluated in each of Tests 1–6. Results Among the three groups, the expert group had the highest average scores on Tests 1–4 (expert: 3.86, 6.57, 7.43, and 7.57; coaching: 1.86, 6.14, 6, and 6.29; and non-coaching: 1.14, 4.14, 4.71, and 4.86). The coaching group’s scores caught up with the expert groups’ average scores on Tests 5 and 6, whereas those of the non-coaching group did not (expert and coaching: 7.86 and 8.00, non-coaching: 5.86 and 7.14). All examinees in the expert and coaching groups achieved a perfect score on the final Test 6, whereas three of the seven non-coaching trainees did not. Conclusion SNIPURS might be suitable for vertebral needle targeting training. The coaching provided during SNIPURS training helped the trainees to acquire the spinal puncture techniques in PVP.
was equal to the GTV. The internal target volume (ITV) was derived from the GTV delineations on a maximum intensity projection. The SBRT was delivery to respiratory movement using image-guided radiotherapy (IGRT). Dose volume histogram in the exhalation phase CT was analyzed using a treatment planning system with an anisotropic analytical algorithm. Tumors were divided into two groups by location: in lung field (lowdensity) and on chest wall (high-density) groups. The volumetric dose differences between the GTV and PTV were assessed in both groups. Local control rates of the doses to the GTV and PTV were compared. Results: Seventy-nine patients' dose-volume histograms were analyzed. The median patient age was 77 (range: 29e95) years (51 male, 28 female patients). Sixty-two and 17 patients received SBRT doses of 50 and 60 Gy respectively, in 5 fractions. The actuarial 3-year local control rate was 80.1% (95% confidence interval: 65.9e88.8), during the median observation period was 25.9 (range: 6.2e106.7) months. All volumetric doses that indicated a difference between the GTV and the PTV were significantly higher in the lung field than on the chest wall (p < 0.01). The homogeneity index (HI) of the PTV in lung field was significantly higher than that on chest wall (p Z 0.03), but the HI differences in the GTV were not significant. Local control was only significantly higher when 50% of the GTV (D50 %) was prescribed 100 Gy 10 (biological equivalent dose, ab Z 10 Gy) and more (p Z 0.008). The local control rates were not significant for the GTV D98 % (p Z 0.06), PTV D98 % (p Z 0.11), or PTV D50 % (p Z 0.06). The GTV D50 % [Hazard ratio (HR) Z 0.25, p Z 0.04] and GTV volume (HR Z 4.26, p Z 0.04) were significant prognostic factors for local control, but the results for PTV D50 % were not significant by multivariate analysis. Conclusion: The dose differences between the GTV and the PTV were significant according to tumor location. The prescribed dose to the GTV had more impact on local control than that to the PTV. A detailed volumetric description of the GTV is needed to achieve optimal local control of stage I NSCLC.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.