Medical imaging call centre: a communication success story.

AIM: To evaluate utilisation of a medical imaging call centre (MICC) at a multi-site, academic radiology department, focusing on communication of critical, urgent, or significant unexpected findings. MATERIALS AND METHODS: Institutional research ethics board approval was obtained. All calls made to MICC from 1 January to 31 December 2019 were reviewed retrospectively. The total number of calls, date, and reason of each call, level of report alert, and turnaround time (TAT) were recorded. Level 1, 2, and 3 alerts were defined as "potentially life-threatening new/unexpected findings", "could result in morbidity/mortality", or "not immediately life-threatening or urgent", respectively. TAT was defined as the time from alert request received by the MICC until acknowledgement of receipt by the referring physician, with a desired TAT of 60 min, 3 h, and 3 days for each level, respectively. RESULTS: The MICC received 29,799 calls in 2019, on average 2,483 (range 1,989-3,098) calls per month. The most common indications for contacting the MICC were to request imaging reports to be expedited (14,916 calls, 50%) and issuing report alerts to communicate unexpected or urgent findings (7,060 calls, 24%). Average number and range of calls for Level 1, 2, and 3 alerts were 57 (39-80), 345 (307-388), and 187 (127-215) per month, respectively. Average TAT for Level 1, 2, and 3 report alerts were 59 min, 2 h 26 min, and 19 h 39 min, respectively. CONCLUSION: The MICC received a large volume of calls and was a successful method for timely communication of unexpected or urgent imaging findings using a three-tiered alert system.

Dynamic Learning Rate in Deep CNN Model for Metastasis Detection and Classification of Histopathology Images.

Diagnosis of different breast cancer stages using histopathology whole slide images (WSI) is the gold standard in determining the grade of tissue metastasis. Computer-aided diagnosis (CAD) assists medical experts as a second opinion tool in early detection to prevent further proliferation. The field of pathology has advanced so rapidly that it is possible to obtain high-quality images from glass slides. Patches from the region of interest in histopathology images are extracted and trained using artificial neural network models. The trained model primarily analyzes and predicts the histology images for the benign or malignant class to which it belongs. Classification of medical images focuses on the training of models with layers of abstraction to distinguish between these two classes with less false-positive rates. The learning rate is the crucial hyperparameter used during the training of deep convolutional neural networks (DCNN) to improve model accuracy. This work emphasizes the relevance of the dynamic learning rate than the fixed learning rate during the training of networks. The dynamic learning rate varies with preset conditions between the lower and upper boundaries and repeats at different iterations. The performance of the model thus improves and attains comparatively high accuracy with fewer iterations.

Multicolor image classification using the multimodal information bottleneck network (MMIB-Net) for detecting diabetic retinopathy.

Multicolor (MC) imaging is an imaging modality that records confocal scanning laser ophthalmoscope (cSLO) fundus images, which can be used for the diabetic retinopathy (DR) detection. By utilizing this imaging technique, multiple modal images can be obtained in a single case. Additional symptomatic features can be obtained if these images are considered during the diagnosis of DR. However, few studies have been carried out to classify MC Images using deep learning methods, let alone using multi modal features for analysis. In this work, we propose a novel model which uses the multimodal information bottleneck network (MMIB-Net) to classify the MC Images for the detection of DR. Our model can extract the features of multiple modalities simultaneously while finding concise feature representations of each modality using the information bottleneck theory. MC Images classification can be achieved by picking up the combined representations and features of all modalities. In our experiments, it is shown that the proposed method can achieve an accurate classification of MC Images. Comparative experiments also demonstrate that the use of multimodality and information bottleneck improves the performance of MC Images classification. To the best of our knowledge, this is the first report of DR identification utilizing the multimodal information bottleneck convolutional neural network in MC Images.

Methodological considerations of evaluating the rate of presynaptic dopaminergic denervation in Parkinson disease with radiotracers: Analysis of the PPMI data.

ABSTRACT: Many previous studies have estimated the rate of dopaminergic denervation in Parkinson disease (PD) via imaging studies. However, they lack the considerations of onset age, disease duration at onset, gender, and dopaminergic denervation due to normal aging. Herein, using a large prospective cohort, we estimated the rate of dopaminergic denervation in PD patients, compared with an age- and gender-matched normal control group.One hundred forty-one normal controls and 301 PD patients were enrolled. Striatal specific binding ratios (SBRs) of I-123 FP-CIT single positron emission tomography images were analyzed according to the age of onset, gender, and the duration of motor symptoms.In the PD group, symptom duration was significantly correlated with caudate SBRs, but with putamen SBRs (P  < .05, R2 = 0.02). Moreover, was significantly inversely related to caudate SBRs, but not with putamen SBRs (P  < .05, R2 = 0.02). Patients of different age onsets did not show any significant correlation between symptom durations and striatal SBRs. In the age-matched group, no significant relationship was observed between symptom duration and percent decrease of caudate SBRs, but there was a significant relationship between symptom duration and percent decrease of the putamen SBRs (P  < .01, R2 = 0.06). There was no significant relationship between the symptom duration and the percent decrease of striatal SBRs in the age- and gender-matched group.The significance and R2 values from the regression analysis between symptom duration, age, and dopaminergic denervation are low. This suggests that, contrary to previous knowledge, there is a relatively weak association between dopaminergic denervation and age or symptom duration.

New types of localization methods for adrenocorticotropic hormone-dependent Cushing's syndrome.

The management of endogenous Cushing's syndrome (CS) typically involves two key steps: (i) confirmation of autonomous hypercortisolism and (ii) localization of the cause to guide treatment. Adrenocorticotropic hormone (ACTH)-dependent CS is most commonly due to a pituitary corticotrope tumor which may be so small as to evade detection on conventional magnetic resonance imaging (MRI). Although biochemical testing (e.g., corticotropin stimulation; dexamethasone suppression) can provide an indication of the likely origin of ACTH excess, bilateral inferior petrosal sinus catheterization offers greater accuracy to distinguish pituitary-driven CS [Cushing's Disease (CD)] from the ectopic ACTH syndrome [EAS, e.g., due to a bronchial or pancreatic neuroendocrine tumor (NET)]. In patients with CD, 40-50% may not have a pituitary adenoma (PA) readily visualized on standard clinical MRI. In these subjects, alternative MR sequences (e.g., dynamic, volumetric, fluid attenuation inversion recovery) and higher magnetic field strength (7T > 3T > 1.5T) may aid tumor localization but carry a risk of identifying coincidental (non-causative) pituitary lesions. Molecular imaging is therefore increasingly being deployed to detect small ACTH-secreting PA, with hybrid imaging [e.g., positron emission tomography (PET) combined with MRI] allowing precise anatomical localization of sites of radiotracer (e.g., 11C-methionine) uptake. Similarly, small ACTH-secreting NETs, missed on initial cross-sectional imaging, may be detected using PET tracers targeting abnormal glucose metabolism (e.g., 18F-fluorodeoxyglucose), somatostatin receptor (SSTR) expression (e.g., 68Ga-DOTATATE), amine precursor (e.g., 18F-DOPA) or amino acid (e.g., 11C-methionine) uptake. Therefore, modern management of ACTH-dependent CS should ideally be undertaken in specialist centers which have an array of cross-sectional and functional imaging techniques at their disposal.

Universal adversarial attacks on deep neural networks for medical image classification.

BACKGROUND: Deep neural networks (DNNs) are widely investigated in medical image classification to achieve automated support for clinical diagnosis. It is necessary to evaluate the robustness of medical DNN tasks against adversarial attacks, as high-stake decision-making will be made based on the diagnosis. Several previous studies have considered simple adversarial attacks. However, the vulnerability of DNNs to more realistic and higher risk attacks, such as universal adversarial perturbation (UAP), which is a single perturbation that can induce DNN failure in most classification tasks has not been evaluated yet. METHODS: We focus on three representative DNN-based medical image classification tasks (i.e., skin cancer, referable diabetic retinopathy, and pneumonia classifications) and investigate their vulnerability to the seven model architectures of UAPs. RESULTS: We demonstrate that DNNs are vulnerable to both nontargeted UAPs, which cause a task failure resulting in an input being assigned an incorrect class, and to targeted UAPs, which cause the DNN to classify an input into a specific class. The almost imperceptible UAPs achieved > 80% success rates for nontargeted and targeted attacks. The vulnerability to UAPs depended very little on the model architecture. Moreover, we discovered that adversarial retraining, which is known to be an effective method for adversarial defenses, increased DNNs' robustness against UAPs in only very few cases. CONCLUSION: Unlike previous assumptions, the results indicate that DNN-based clinical diagnosis is easier to deceive because of adversarial attacks. Adversaries can cause failed diagnoses at lower costs (e.g., without consideration of data distribution); moreover, they can affect the diagnosis. The effects of adversarial defenses may not be limited. Our findings emphasize that more careful consideration is required in developing DNNs for medical imaging and their practical applications.

Artificial Intelligence and Its Effect on Dermatologists' Accuracy in Dermoscopic Melanoma Image Classification: Web-Based Survey Study.

BACKGROUND: Early detection of melanoma can be lifesaving but this remains a challenge. Recent diagnostic studies have revealed the superiority of artificial intelligence (AI) in classifying dermoscopic images of melanoma and nevi, concluding that these algorithms should assist a dermatologist's diagnoses. OBJECTIVE: The aim of this study was to investigate whether AI support improves the accuracy and overall diagnostic performance of dermatologists in the dichotomous image-based discrimination between melanoma and nevus. METHODS: Twelve board-certified dermatologists were presented disjoint sets of 100 unique dermoscopic images of melanomas and nevi (total of 1200 unique images), and they had to classify the images based on personal experience alone (part I) and with the support of a trained convolutional neural network (CNN, part II). Additionally, dermatologists were asked to rate their confidence in their final decision for each image. RESULTS: While the mean specificity of the dermatologists based on personal experience alone remained almost unchanged (70.6% vs 72.4%; P=.54) with AI support, the mean sensitivity and mean accuracy increased significantly (59.4% vs 74.6%; P=.003 and 65.0% vs 73.6%; P=.002, respectively) with AI support. Out of the 10% (10/94; 95% CI 8.4%-11.8%) of cases where dermatologists were correct and AI was incorrect, dermatologists on average changed to the incorrect answer for 39% (4/10; 95% CI 23.2%-55.6%) of cases. When dermatologists were incorrect and AI was correct (25/94, 27%; 95% CI 24.0%-30.1%), dermatologists changed their answers to the correct answer for 46% (11/25; 95% CI 33.1%-58.4%) of cases. Additionally, the dermatologists' average confidence in their decisions increased when the CNN confirmed their decision and decreased when the CNN disagreed, even when the dermatologists were correct. Reported values are based on the mean of all participants. Whenever absolute values are shown, the denominator and numerator are approximations as every dermatologist ended up rating a varying number of images due to a quality control step. CONCLUSIONS: The findings of our study show that AI support can improve the overall accuracy of the dermatologists in the dichotomous image-based discrimination between melanoma and nevus. This supports the argument for AI-based tools to aid clinicians in skin lesion classification and provides a rationale for studies of such classifiers in real-life settings, wherein clinicians can integrate additional information such as patient age and medical history into their decisions.

Diagnostic Accuracy of Intraoperative Tools for Detecting Endometriosis: A Systematic Review and Meta-analysis.

OBJECTIVE: To evaluate the diagnostic accuracy of intraoperative laparoscopic imaging tools in reference to that of histopathology for detecting endometriotic lesions and to compare them with conventional white-light inspection by performing a systematic review with meta-analysis. DATA SOURCES: We searched the MEDLINE, EMBASE, and CENTRAL databases in addition to citations and reference lists until the end of February 2019. METHODS OF STUDY SELECTION: Two authors screened 1038 citations for eligibility. We included randomized controlled trials or prospective cohort studies published in English, assessing the accuracy of intraoperative imaging tools for diagnosing endometriosis during laparoscopy. We considered studies using histopathologic evaluation as a standard criterion. TABULATION, INTEGRATION, AND RESULTS: Seven studies were eligible, including 472 women and 1717 histopathologic specimens, and they involved study of the use of narrow-band imaging (2 studies), 5-aminolevulinic acid-induced fluorescence (2 studies), autofluorescence imaging (1 study), indocyanine green (1 study), and a 3-dimensional robotic laparoscopy (1 study). Two authors extracted data and assessed the validity of the included studies. Bivariate random-effects models and McNemar's test were used to compare the tests and evaluate sources of heterogeneity. Four studies were attributed a high risk of bias, and biopsies of normal-looking peritoneum were not performed to verify the results in 3 studies; both factors were identified as significant sources of heterogeneity, leading to the overestimation of the sensitivity and underestimation of the specificity of imaging tools. In all studies, additional endometriotic lesions were diagnosed with the enhanced imaging tool compared with white-light inspection alone. In the 4 studies that appropriately performed control biopsies (171 women, 448 specimens), enhanced imaging techniques were associated with a higher sensitivity and specificity compared with white-light inspection (0.84 and 0.89 compared with 0.75 and 0.76, respectively, p ≤.001). Adverse events were uncommon (n = 5) and reported only with the use of exogeneous photosensitizers. There were no reports of long-term changes in patient-reported outcomes arising from better detection of endometriosis lesions. CONCLUSION: Studies report that enhanced imaging allows for the detection of additional endometriotic lesions missed by conventional white-light laparoscopy. The benefits of finding these additional lesions using enhanced imaging compared with white-light inspection alone on long-term postoperative outcomes have not been determined, and these tools should be considered only in a research context at this time.

Trends in Medical Imaging During Pregnancy in the United States and Ontario, Canada, 1996 to 2016.

Importance: The use of medical imaging has sharply increased over the last 2 decades. Imaging rates during pregnancy have not been quantified in a large, multisite study setting. Objective: To evaluate patterns of medical imaging during pregnancy. Design, Setting, and Participants: A retrospective cohort study was performed at 6 US integrated health care systems and in Ontario, Canada. Participants included pregnant women who gave birth to a live neonate of at least 24 weeks' gestation between January 1, 1996, and December 31, 2016, and who were enrolled in the health care system for the entire pregnancy. Exposures: Computed tomography (CT), magnetic resonance imaging, conventional radiography, angiography and fluoroscopy, and nuclear medicine. Main Outcomes and Measures: Imaging rates per pregnancy stratified by country and year of child's birth. Results: A total of 3 497 603 pregnancies in 2 211 789 women were included. Overall, 26% of pregnancies were from US sites. Most (92%) were in women aged 20 to 39 years, and 85% resulted in full-term births. Computed tomography imaging rates in the United States increased from 2.0 examinations/1000 pregnancies in 1996 to 11.4/1000 pregnancies in 2007, remained stable through 2010, and decreased to 9.3/1000 pregnancies by 2016, for an overall increase of 3.7-fold. Computed tomography rates in Ontario, Canada, increased more gradually by 2.0-fold, from 2.0/1000 pregnancies in 1996 to 6.2/1000 pregnancies in 2016, which was 33% lower than in the United States. Overall, 5.3% of pregnant women in US sites and 3.6% in Ontario underwent imaging with ionizing radiation, and 0.8% of women at US sites and 0.4% in Ontario underwent CT. Magnetic resonance imaging rates increased steadily from 1.0/1000 pregnancies in 1996 to 11.9/1000 pregnancies in 2016 in the United States and from 0.5/1000 pregnancies in 1996 to 9.8/1000 pregnancies in 2016 in Ontario, surpassing CT rates in 2013 in the United States and in 2007 in Ontario. In the United States, radiography rates doubled from 34.5/1000 pregnancies in 1996 to 72.6/1000 pregnancies in 1999 and then decreased to 47.6/1000 pregnancies in 2016; rates in Ontario slowly increased from 36.2/1000 pregnancies in 1996 to 44.7/1000 pregnancies in 2016. Angiography and fluoroscopy and nuclear medicine use rates were low (5.2/1000 pregnancies), but in most years, higher in Ontario than the United States. Imaging rates were highest for women who were younger than 20 years or aged 40 years or older, gave birth preterm, or were black, Native American, or Hispanic (US data only). Considering advanced imaging only, chest imaging of pregnant women was more likely to use CT in the United States and nuclear medicine imaging in Ontario. Conclusions and Relevance: The use of CT during pregnancy substantially increased in the United States and Ontario over the past 2 decades. Imaging rates during pregnancy should be monitored to avoid unnecessary exposure of women and fetuses to ionizing radiation.

Assessing liver fibrosis without biopsy in patients with HCV or NAFLD.

Staging of liver fibrosis is increasingly done using noninvasive methods, in some cases obviating the need for liver biopsy. Scores based on laboratory values and demographic variables have been developed and validated for assessing fibrosis in patients with hepatitis C virus (HCV) infection and nonalcoholic fatty liver disease (NAFLD), as have several imaging methods that measure shear-wave velocity, a reflection of fibrosis severity.

Swedish Society of Rheumatology 2018 guidelines for investigation, treatment, and follow-up of giant cell arteritis.

Objective: To develop evidence-based guidelines for the management of giant cell arteritis (GCA) as a complement to guidelines in other areas of rheumatology, issued by the Swedish Society of Rheumatology. Methods: A working group selected key areas for recommendations, reviewed the available evidence, and wrote draft guidelines. These were discussed and revised according to standard procedures within the Swedish Society of Rheumatology, including a one-day meeting open to all members. For key recommendations, the quality of evidence was assessed according to GRADE. The final guidelines were approved by the Society board in March 2018. Results: The guidelines include recommendations on diagnostic procedures, pharmacological treatment, follow-up, and adjuvant treatment. Ultrasonography is complementary to temporal artery biopsy (TAB) in the diagnostic work-up. Other imaging techniques (magnetic resonance imaging and positron emission tomography/computed tomography) are important in evaluating large-vessel involvement. Glucocorticoids (oral, or intravenous in cases with ischaemic complications) remain the first line treatment for GCA. Addition of tocilizumab is recommended for patients with relapsing disease who meet five criteria, representing active disease that has been objectively verified by TAB or imaging. Tocilizumab may also be considered in patients with newly diagnosed GCA who are at major risk of severe glucocorticoid side effects. Based on current evidence, tocilizumab treatment for > 1 year cannot be recommended. Conclusion: These guidelines are based on current evidence and consensus within Swedish rheumatology. Following major developments in diagnostics and treatment of GCA, such guidelines are important for clinical practice, and should be updated on a regular basis.

Plus Disease in Telemedicine Approaches to Evaluating Acute-Phase ROP (e-ROP) Study: Characteristics, Predictors, and Accuracy of Image Grading.

PURPOSE: To describe characteristics and predictors of plus disease, and the accuracy of image grading for plus disease in the e-ROP Study. DESIGN: Secondary analyses of data from 13 North American centers. PARTICIPANTS: Premature infants with birth weight (BW) <1251 g. METHODS: Infants underwent regularly scheduled diagnostic examinations by ophthalmologists and digital imaging by trained imagers using a wide-field digital camera. Two masked nonphysician trained readers independently evaluated images for posterior pole abnormality (normal, preplus, plus), with discrepancies adjudicated by a reading supervisor. Logistic regression models were used to determine predictors for plus disease. The sensitivity and specificity of image grading for plus disease were calculated using the clinical examination finding as reference standard. MAIN OUTCOME MEASURES: Odds ratios (OR), sensitivity, and specificity. RESULTS: Among 1239 infants (mean BW 864 g, mean gestational age [GA] 27 weeks), 129 infants (10%) (226 eyes, 75% bilateral) had plus disease from clinical examination. When plus disease was first diagnosed in clinical examination at median postmenstrual age (PMA) of 36 weeks (range: 32-43 weeks), 94% to 96% of plus occurred in the superior or inferior temporal quadrant. Significant predictors for plus disease from multivariate analysis were as follows: GA (OR = 3.2 for ≤24 vs. ≥28 weeks, P = 0.004), race (OR = 2.4 for white vs. black, P = 0.01), respiratory support (OR = 7.1, P = 0.006), weight gain (OR = 1.5 for weight gain ≤12 vs. >18 g/day, P = 0.03), and image findings at the first image session, including presence of preplus/plus disease (OR = 2.7, P = 0.003), ROP stage (OR = 4.2 for stage 3 ROP vs. no ROP, P = 0.006), and blot hemorrhage (OR = 3.1, P = 0.003). These features predicted plus disease with an area under the receiver operating characteristic curve of 0.89 (95% confidence interval [CI]: 0.85-0.92). The image grading using preplus as the cut point had sensitivity of 94% (95% CI: 90%-97%) and specificity of 81% (95% CI: 79%-82%) for detecting plus disease in an eye. CONCLUSIONS: Among e-ROP infants, plus disease developed in 10% of infants at a median PMA of 37 weeks, with the majority being bilateral and mostly in the superior or inferior temporal quadrant. GA, race, respiratory support, postnatal weight gain, image findings of the posterior pole, and ROP predict development of plus disease. Nonphysician image grading can detect almost all plus disease with good specificity.

A novel fused convolutional neural network for biomedical image classification.

With the advent of biomedical imaging technology, the number of captured and stored biomedical images is rapidly increasing day by day in hospitals, imaging laboratories and biomedical institutions. Therefore, more robust biomedical image analysis technology is needed to meet the requirement of the diagnosis and classification of various kinds of diseases using biomedical images. However, the current biomedical image classification methods and general non-biomedical image classifiers cannot extract more compact biomedical image features or capture the tiny differences between similar images with different types of diseases from the same category. In this paper, we propose a novel fused convolutional neural network to develop a more accurate and highly efficient classifier for biomedical images, which combines shallow layer features and deep layer features from the proposed deep neural network architecture. In the analysis, it was observed that the shallow layers provided more detailed local features, which could distinguish different diseases in the same category, while the deep layers could convey more high-level semantic information used to classify the diseases among the various categories. A detailed comparison of our approach with traditional classification algorithms and popular deep classifiers across several public biomedical image datasets showed the superior performance of our proposed method for biomedical image classification. In addition, we also evaluated the performance of our method in modality classification of medical images using the ImageCLEFmed dataset. Graphical abstract The graphical abstract shows the fused, deep convolutional neural network architecture proposed for biomedical image classification. In the architecture, we can clearly see the feature-fusing process going from shallow layers and the deep layers.

The "National Integrated Medical Imaging System" [NIMIS]-friend, not nimesis!

INTRODUCTION: The "National Integrated Medical Imaging System" or NIMIS went live in 2011 and allows the movement of patient radiology imaging throughout the Irish health system. At the time of its launch, NIMIS was not only going to allow the filmless passage of patient radiology imaging but it was also envisaged that it would act as a medical image archive. The aim of this study was to assess the awareness and use of non-consultant hospital doctors and hospital consultants with regard to this medical image archive/referral function of NIMIS. METHODS: A survey was carried out on 50 doctors across all specialities and grades at Tullamore Hospital looking at different aspects of the use of NIMIS. RESULTS: Ninety-four percent of respondents use NIMIS on a daily basis and 6% use it on a weekly basis. The primary reason for using NIMIS was found to be "Viewing and Ordering Imaging" in 92% of those surveyed with 8% stating it was "Viewing imaging/reports". Ninety-eight percent surveyed said they had never used NIMIS to send a referral form or clinical photograph and 82% were not aware of this potential function. The majority of those surveyed stated that they either agreed or strongly agreed NIMIS is user-friendly. CONCLUSION: NIMIS allows the safe and confidential flow of patient images and clinical information in the Irish health system. It could provide definite potential in the areas of clinical conferencing, multidisciplinary meetings and remote patient assessment along with collaborative research and education.

Why rankings of biomedical image analysis competitions should be interpreted with care.

International challenges have become the standard for validation of biomedical image analysis methods. Given their scientific impact, it is surprising that a critical analysis of common practices related to the organization of challenges has not yet been performed. In this paper, we present a comprehensive analysis of biomedical image analysis challenges conducted up to now. We demonstrate the importance of challenges and show that the lack of quality control has critical consequences. First, reproducibility and interpretation of the results is often hampered as only a fraction of relevant information is typically provided. Second, the rank of an algorithm is generally not robust to a number of variables such as the test data used for validation, the ranking scheme applied and the observers that make the reference annotations. To overcome these problems, we recommend best practice guidelines and define open research questions to be addressed in the future.

Comparison of medical image classification accuracy among three machine learning methods.

BACKGROUND: Low-quality medical images may influence the accuracy of the machine learning process. OBJECTIVE: This study was undertaken to compare accuracy of medical image classification among machine learning methods, as classification is a basic aspect of clinical image inspection. METHODS: Three types of machine learning methods were used, which include Support Vector Machine (SVM), Artificial Neural Network (ANN), and Convolution Neural Network (CNN). To investigate changes in accuracy related to image quality, we constructed a single dataset using two different file formats of DICOM (Digital Imaging and Communications in Medicine) and JPEG (Joint Photographic Experts Group). RESULTS: The JPEG format contains less color information and data capacity than the DICOM format. CNN classification was accurate for both datasets, whereas SVM and ANN accuracy decreased with the loss of data from DICOM to JPEG formats. CONCLUSIONS: CNN is more accurate than conventional machine learning methods that utilize the manual feature extraction.

Classification of Medical Images in the Biomedical Literature by Jointly Using Deep and Handcrafted Visual Features.

The classification of medical images and illustrations from the biomedical literature is important for automated literature review, retrieval, and mining. Although deep learning is effective for large-scale image classification, it may not be the optimal choice for this task as there is only a small training dataset. We propose a combined deep and handcrafted visual feature (CDHVF) based algorithm that uses features learned by three fine-tuned and pretrained deep convolutional neural networks (DCNNs) and two handcrafted descriptors in a joint approach. We evaluated the CDHVF algorithm on the ImageCLEF 2016 Subfigure Classification dataset and it achieved an accuracy of 85.47%, which is higher than the best performance of other purely visual approaches listed in the challenge leaderboard. Our results indicate that handcrafted features complement the image representation learned by DCNNs on small training datasets and improve accuracy in certain medical image classification problems.

Representing Medical Images With Encoded Local Projections.

This paper introduces the "encoded local projections" (ELP) as a new dense-sampling image descriptor for search and classification problems. The gradient changes of multiple projections in local windows of gray-level images are encoded to build a histogram that captures spatial projection patterns. Using projections is a conventional technique in both medical imaging and computer vision. Furthermore, powerful dense-sampling methods, such as local binary patterns and the histogram of oriented gradients, are widely used for image classification and recognition. Inspired by many achievements of such existing descriptors, we explore the design of a new class of histogram-based descriptors with particular applications in medical imaging. We experiment with three public datasets (IRMA, Kimia Path24, and CT Emphysema) to comparatively evaluate the performance of ELP histograms. In light of the tremendous success of deep architectures, we also compare the results with deep features generated by pretrained networks. The results are quite encouraging as the ELP descriptor can surpass both conventional and deep descriptors in performance in several experimental settings.

Collocation for Diffeomorphic Deformations in Medical Image Registration.

Diffeomorphic deformation is a popular choice in medical image registration. A fundamental property of diffeomorphisms is invertibility, implying that once the relation between two points A to B is found, then the relation B to A is given per definition. Consistency is a measure of a numerical algorithm's ability to mimic this invertibility, and achieving consistency has proven to be a challenge for many state-of-the-art algorithms. We present CDD (Collocation for Diffeomorphic Deformations), a numerical solution to diffeomorphic image registration, which solves for the Stationary Velocity Field (SVF) using an implicit A-stable collocation method. CDD guarantees the preservation of the diffeomorphic properties at all discrete points and is thereby consistent to machine precision. We compared CDD's collocation method with the following standard methods: Scaling and Squaring, Forward Euler, and Runge-Kutta 4, and found that CDD is up to 9 orders of magnitude more consistent. Finally, we evaluated CDD on a number of standard bench-mark data sets and compared the results with current state-of-the-art methods: SPM-DARTEL, Diffeomorphic Demons and SyN. We found that CDD outperforms state-of-the-art methods in consistency and delivers comparable or superior registration precision.

Breast cancer diagnosis: Imaging techniques and biochemical markers.

Breast cancer is a complex disease which is found as the second cause of cancer-associated death among women. Accumulating of evidence indicated that various factors (i.e., gentical and envirmental factors) could be associated with initiation and progression of breast cancer. Diagnosis of breast cancer patients in early stages is one of important aspects of breast cancer treatment. Among of various diagnosis platforms, imaging techniques are main diagnosis approaches which could provide valuable data on patients with breast cancer. It has been showed that various imaging techniques such as mammography, magnetic resonance imaging (MRI), positron-emission tomography (PET), Computed tomography (CT), and single-photon emission computed tomography (SPECT) could be used for diagnosis and monitoring patients with breast cancer in various stages. Beside, imaging techniques, utilization of biochemical biomarkers such as proteins, DNAs, mRNAs, and microRNAs could be employed as new diagnosis and therapeutic tools for patients with breast cancer. Here, we summarized various imaging techniques and biochemical biomarkers could be utilized as diagnosis of patients with breast cancer. Moreover, we highlighted microRNAs and exosomes as new diagnosis and therapeutic biomarkers for monitoring patients with breast cancer.