scholarly journals Assessment of deep neural networks for the diagnosis of benign and malignant skin neoplasms in comparison with dermatologists: A retrospective validation study

PLoS Medicine ◽  
2020 ◽  
Vol 17 (11) ◽  
pp. e1003381
Author(s):  
Seung Seog Han ◽  
Ik Jun Moon ◽  
Seong Hwan Kim ◽  
Jung-Im Na ◽  
Myoung Shin Kim ◽  
...  
Keyword(s):  
Neural Networks ◽  
Sensitivity And Specificity ◽  
Real World ◽  
Clinical Information ◽  
Skin Neoplasms ◽  
Validation Dataset ◽  
Clinical Impression ◽  
Clinical Images ◽  
Clinical Diagnoses ◽  
Attending Physicians

Background The diagnostic performance of convolutional neural networks (CNNs) for diagnosing several types of skin neoplasms has been demonstrated as comparable with that of dermatologists using clinical photography. However, the generalizability should be demonstrated using a large-scale external dataset that includes most types of skin neoplasms. In this study, the performance of a neural network algorithm was compared with that of dermatologists in both real-world practice and experimental settings. Methods and findings To demonstrate generalizability, the skin cancer detection algorithm (https://rcnn.modelderm.com) developed in our previous study was used without modification. We conducted a retrospective study with all single lesion biopsied cases (43 disorders; 40,331 clinical images from 10,426 cases: 1,222 malignant cases and 9,204 benign cases); mean age (standard deviation [SD], 52.1 [18.3]; 4,701 men [45.1%]) were obtained from the Department of Dermatology, Severance Hospital in Seoul, Korea between January 1, 2008 and March 31, 2019. Using the external validation dataset, the predictions of the algorithm were compared with the clinical diagnoses of 65 attending physicians who had recorded the clinical diagnoses with thorough examinations in real-world practice. In addition, the results obtained by the algorithm for the data of randomly selected batches of 30 patients were compared with those obtained by 44 dermatologists in experimental settings; the dermatologists were only provided with multiple images of each lesion, without clinical information. With regard to the determination of malignancy, the area under the curve (AUC) achieved by the algorithm was 0.863 (95% confidence interval [CI] 0.852–0.875), when unprocessed clinical photographs were used. The sensitivity and specificity of the algorithm at the predefined high-specificity threshold were 62.7% (95% CI 59.9–65.1) and 90.0% (95% CI 89.4–90.6), respectively. Furthermore, the sensitivity and specificity of the first clinical impression of 65 attending physicians were 70.2% and 95.6%, respectively, which were superior to those of the algorithm (McNemar test; p < 0.0001). The positive and negative predictive values of the algorithm were 45.4% (CI 43.7–47.3) and 94.8% (CI 94.4–95.2), respectively, whereas those of the first clinical impression were 68.1% and 96.0%, respectively. In the reader test conducted using images corresponding to batches of 30 patients, the sensitivity and specificity of the algorithm at the predefined threshold were 66.9% (95% CI 57.7–76.0) and 87.4% (95% CI 82.5–92.2), respectively. Furthermore, the sensitivity and specificity derived from the first impression of 44 of the participants were 65.8% (95% CI 55.7–75.9) and 85.7% (95% CI 82.4–88.9), respectively, which are values comparable with those of the algorithm (Wilcoxon signed-rank test; p = 0.607 and 0.097). Limitations of this study include the exclusive use of high-quality clinical photographs taken in hospitals and the lack of ethnic diversity in the study population. Conclusions Our algorithm could diagnose skin tumors with nearly the same accuracy as a dermatologist when the diagnosis was performed solely with photographs. However, as a result of limited data relevancy, the performance was inferior to that of actual medical examination. To achieve more accurate predictive diagnoses, clinical information should be integrated with imaging information.

scholarly journals Retrospective Assessment of Deep Neural Networks for Skin Tumor Diagnosis

2019 ◽  
Author(s):  
Seung Seog Han ◽  
Ik Jun Moon ◽  
Jung-Im Na ◽  
Myoung Shin Kim ◽  
Gyeong Hun Park ◽  
...  
Keyword(s):  
Clinical Diagnosis ◽  
External Validation ◽  
Skin Tumors ◽  
Validation Dataset ◽  
Rank Test ◽  
Clinical Images ◽  
Task Classification ◽  
Malignant Skin Tumors ◽  
Sensitivity Specificity ◽  
Attending Physicians

ABSTRACTBACKGROUNDThe aim of this study was to validate the performance of algorithm (http://rcnn.modelderm.com) for the diagnosis of benign and malignant skin tumors.METHODSWith external validation dataset (43 disorders; 40,331 clinical images from 10,426 patients; January 1, 2008 – March 31, 2019), we compared the prediction of algorithm with the clinical diagnosis of 65 attending physicians at the time of biopsy request.RESULTSFor binary-task classification of determining malignancy, the AUC of the algorithm was 0.863(95% CI 0.852-0.875) with unprocessed clinical photographs. The sensitivity/specificity of the algorithm at the predefined high-sensitivity and high-specificity threshold were 79.1%(76.9-81.4)/76.9%(76.1-77.8) and 62.7%(59.9- 65.5)/90.0%(89.4-90.6), respectively. The sensitivity/specificity calculated by the clinical diagnosis of attending physicians were 88.1%/83.8%(Top-3) and 70.2%/95.6%(Top-1), which were superior to those of algorithm.For multi-task classification, the mean Top-1,2,3 accuracies of the algorithm were 42.6±20.7%, 56.1±22.8%, 61.9±22.9%, and those of clinical diagnosis were 65.4±17.7%, 73.9±16.6%, 74.7±16.6%, respectively.In the reader test with images from 30-patients batches, the sensitivity / specificity of the algorithm at the predefined threshold were 66.9%±30.2% / 87.4%±16.5%. The sensitivity / specificity derived from the first diagnosis of 44 the participants were 65.8%±33.3% / 85.7%±11.0%, which were comparable with those of the algorithm (Wilcoxon signed-rank test; P=0.61 / 0.097).CONCLUSIONSOur algorithm could diagnose skin tumors at dermatologist-level when diagnosis was made solely with photographs, demonstrating its potential as a mass screening tool in telemedicine setting. However, due to limited data relevancy, the performance was inferior to that of actual medical examination. Clinical information should be integrated with imaging information to achieve more accurate predictions.

scholarly journals 332 The diagnostic role of shunt series radiographs (SSR) in children presenting to the children’s emergency department (CED) with suspected ventriculoperitoneal (VP) shunt failure

2020 ◽  
Vol 37 (12) ◽  
pp. 852.3-853
Author(s):  
Angharad Griffiths ◽  
Ikechukwu Okafor ◽  
Thomas Beattie
Keyword(s):  
Clinical Outcome ◽  
Sensitivity And Specificity ◽  
Clinical Information ◽  
Flow Diagram ◽  
List Type ◽  
Shunt Failure ◽  
Vp Shunt ◽  
Wide Range ◽  
Diagnostic Role

Aims/Objectives/BackgroundVP shunts are used to drain CSF from the cranial vault because of a wide range of pathologies and, like any piece of hardware, can fail. Traditionally investigations include SSR and CT. This project examines the role of SSR in evaluating children with suspected VP shunt failure.Primary outcome: Sensitivity and specificity of SSR in children presenting to the CED with concern for shunt failure.Methods/DesignConducted in a single centre, tertiary CED of the national Irish Neurosurgical(NS) referral centre (ED attendance:>50,000 patients/year). 100 sequential SSR requested by the CED were reviewed. Clinical information was extracted from electronic requests. Shunt failure was defined by the need for NS intervention(Revision).Abstract 332 Figure 1Abstract 332 Figure 2Results/ConclusionsSensitivity and specificity is presented in figure 1 (two by two table).100 radiographs performed in 84 children.22% shunts revised (see flow diagram).7 SSR’s were abnormal.85% (n=6) shunts revised. [5 following abnormal CT].Of the normal SSR’s; 16 had abnormal CT and revised.85/100 received CT.64 of 85 CT’s (75%) were normal.□6 of the 64 had focal shunt concern.SSR’s shouldn’t be used in isolation. NPV&PPV, Sensitivity&Specificity is low.SSR’s are beneficial where there’s concern over focal shunt problems (injury/pain/swelling) or following abnormal CT.VP shunt failure is not well investigated with SSR alone.SSR’s could be omitted where there is no focal shunt concern/after normal CT (without impacting clinical outcome) reducing radiation exposure and reduce impact on CED’s.59 SSR’s could have been avoided without adverse clinical outcome.

scholarly journals Deep learning-enabled medical computer vision

2021 ◽  
Vol 4 (1) ◽  
Author(s):  
Andre Esteva ◽  
Katherine Chou ◽  
Serena Yeung ◽  
Nikhil Naik ◽  
Ali Madani ◽  
...  
Keyword(s):  
Artificial Intelligence ◽  
Neural Networks ◽  
Computer Vision ◽  
Deep Learning ◽  
Medical Imaging ◽  
Real World ◽  
Recent Progress ◽  
Medical Applications ◽  
Modern Computer ◽  
Medical Computer

AbstractA decade of unprecedented progress in artificial intelligence (AI) has demonstrated the potential for many fields—including medicine—to benefit from the insights that AI techniques can extract from data. Here we survey recent progress in the development of modern computer vision techniques—powered by deep learning—for medical applications, focusing on medical imaging, medical video, and clinical deployment. We start by briefly summarizing a decade of progress in convolutional neural networks, including the vision tasks they enable, in the context of healthcare. Next, we discuss several example medical imaging applications that stand to benefit—including cardiology, pathology, dermatology, ophthalmology–and propose new avenues for continued work. We then expand into general medical video, highlighting ways in which clinical workflows can integrate computer vision to enhance care. Finally, we discuss the challenges and hurdles required for real-world clinical deployment of these technologies.

An ultrasensitive method for noninvasive pan-cancer early detection based on targeted methylation sequencing of cell-free DNA.

2021 ◽  
Vol 39 (15_suppl) ◽  
pp. 10544-10544
Author(s):  
Tiancheng Han ◽  
Yuanyuan Hong ◽  
Pei Zhihua ◽  
Song Xiaofeng ◽  
Jianing Yu ◽  
...  
Keyword(s):  
Cancer Patients ◽  
Cancer Detection ◽  
Real World ◽  
Early Stage ◽  
Validation Dataset ◽  
Cpg Sites ◽  
Free Dna ◽  
Cancer Types ◽  
Discovery Dataset ◽  
Pan Cancer

10544 Background: Screening the biomarkers from the cell-free DNA (cfDNA) of peripheral blood is a non-invasive and promising method for cancer diagnosis. Among diverse types of biomarkers, epigenetic biomarkers have been reported to be one of the most promising ones. Epigenetic modifications are widespread on the human genome and generally have strong signals due to the similar methylation patterns shared by adjacent CpG sites. Although some epigenetic diagnostic methods have been developed based on cfDNAs, few of them could be applied to pan-cancer and their sensitivities are barely satisfactory for early cancer detection. Methods: Targeted methylation sequencing was performed using our in-house-designed panel targeting regions with abundant cancer-specific methylation CpGs. The cfDNA samples from 80 healthy individuals and 549 cancer patients of 14 cancer types were separately sequenced. The dataset was randomly split into one discovery dataset and one validation dataset. Moreover, cfDNA samples from four cancer patients were diluted with the healthy cfDNAs to generate 12 in vitro simulated samples with low circulating tumor DNA (ctDNA) fraction. Additionally, DNAs extracted from 130 unmatched tumor formalin fixation and paraffin embedding (FFPE) samples of 10 cancer types were sequenced to screen the diagnostic biomarkers. Adjacent CpG sites were first merged into methylation-correlated blocks (MCB) according to their correlations of methylation levels in tumor DNAs. The MCBs with higher methylation levels in tumor DNAs than that of healthy cfDNAs (from the discovery dataset) were defined as our hypermethylation biomarkers. For each cfDNA sample, a hypermethylation score (HM-score) was computed to measure the overall methylation level difference of selected biomarkers. The performance of our method was evaluated with the real-world dataset, while the limit of detection was estimated using the simulated low-ctDNA samples. Results: Our model based on 37 hypermethylation MCB biomarkers achieved an area under the curve (AUC) of 0.89 and 0.86 in the real-world pan-cancer discovery and validation cfDNA datasets, respectively. Furthermore, the overall specificity and sensitivity are 100% and 76.19% in the discovery dataset, and 96.67% and 72.86% in the validation dataset. In the validation dataset, 28/40 (70%) of early-stage colorectal cancer patients and 10/20 (50%) of non-small-cell lung cancer patients were successfully diagnosed. Additionally, all the simulated samples with theoretical ctDNA factions over 0.5% were predicted as diseased, demonstrating the ability of our method to detect tumor signals at early stages. Conclusions: Our cfDNA-based epigenetic method outperforms currently available methods in various cancer types, and is promising to be applied to early-stage cancer detection and samples with low ctDNA fractions.

scholarly journals Real World of Artificial Intelligence - A Review

2020 ◽  
Vol 11 (2) ◽  
pp. 41-47
Author(s):  
Amandeep Kaur ◽  
Madhu Dhiman ◽  
Mansi Tonk ◽  
Ramneet Kaur
Keyword(s):  
Artificial Intelligence ◽  
Neural Networks ◽  
Natural Language Processing ◽  
Face Recognition ◽  
Language Processing ◽  
Real World ◽  
Review Paper ◽  
General Concept ◽  
Research Trends ◽  
Human Intelligence

Artificial Intelligence is the combination of machine and human intelligence, which are in research trends from the last many years. Different Artificial Intelligence programs have become capable of challenging humans by providing Expert Systems, Neural Networks, Robotics, Natural Language Processing, Face Recognition and Speech Recognition. Artificial Intelligence brings a bright future for different technical inventions in various fields. This review paper shows the general concept of Artificial Intelligence and presents an impact of Artificial Intelligence in the present and future world.

scholarly journals Ранняя диагностика меланомы кожи с применением нескольких изображающих систем

2020 ◽  
Vol 128 (6) ◽  
pp. 820
Author(s):  
К.Г. Кудрин ◽  
Е.Н. Римская ◽  
И.А. Аполлонова ◽  
А.П. Николаев ◽  
Н.В. Черномырдин ◽  
...  
Keyword(s):  
Image Processing ◽  
Measurement Errors ◽  
Skin Neoplasms ◽  
Imaging Systems ◽  
Clinical Parameters ◽  
Skin Melanoma ◽  
Clinical Images ◽  
Melanoma Diagnosis ◽  
Complex Approach ◽  
Automated Image Processing

A complex approach to the early diagnosis of skin melanoma has been proposed. The approach involves a step-by-step examination of pigment tumors using several imaging systems. The features of morphometry of clinical images of pigmented skin neoplasms, features of imaging systems, the main stages of automated image processing and pattern recognition in the melanoma diagnosis has been considered. The metrological features of the proposed approach has been shown: the measurement errors of the clinical parameters of skin neoplasms by the proposed methods do not exceed the allowable errors. The approbation of the offered approach has been showed that sensitivity and specificity of the used methods exceeds 90%

scholarly journals INDUSTRIAL APPLICATIONS OF ARTIFICIAL NEURAL NETWORKS

2014 ◽  
pp. 8-20
Author(s):  
Kurosh Madani
Keyword(s):  
Neural Networks ◽  
Artificial Neural Networks ◽  
Real World ◽  
Industrial Applications ◽  
Large Field ◽  
Nonlinear Functions ◽  
The Past ◽  
Real World Applications ◽  
Ann Models ◽  
Artificial Neural

In a large number of real world dilemmas and related applications the modeling of complex behavior is the central point. Over the past decades, new approaches based on Artificial Neural Networks (ANN) have been proposed to solve problems related to optimization, modeling, decision making, classification, data mining or nonlinear functions (behavior) approximation. Inspired from biological nervous systems and brain structure, Artificial Neural Networks could be seen as information processing systems, which allow elaboration of many original techniques covering a large field of applications. Among their most appealing properties, one can quote their learning and generalization capabilities. The main goal of this paper is to present, through some of main ANN models and based techniques, their real application capability in real world industrial dilemmas. Several examples through industrial and real world applications have been presented and discussed.

Abstract 125: Automated External Defibrillator Shock Advisement Performance Among Multiple Human Electrocardiographic Rhythms and Devices

Circulation ◽  
2019 ◽  
Vol 140 (Suppl_2) ◽  
Author(s):  
Xavier J Szigethy ◽  
Connor J Willson ◽  
David D Salcido ◽  
Dylan A Defilippi ◽  
James J Menegazzi
Keyword(s):  
Sensitivity And Specificity ◽  
Real World ◽  
Peak Frequency ◽  
Lead Wire ◽  
Digital To Analog Converter ◽  
Automated External Defibrillators ◽  
Signal Characteristics ◽  
Rhythm Analysis ◽  
High Degree ◽  
Median Slope

Background: Automated external defibrillators (AEDs) perform rhythm analysis in order to facilitate defibrillation. The effectiveness of AEDs is dependent on the accuracy of their rhythm classification, which includes differentiation of shockable rhythms from non-shockable rhythms Independent (i.e. non-industry) evaluation of the performance of AEDs against real-world ECG could lead to improvements in their performance. Objective: To evaluate the sensitivity and specificity characteristics of commercial AEDs with respect to quantitative properties of the ECG waveform in several rhythm presentations using real world ECG data. Methods: We conducted a prospective simulation study evaluating three commercially available AEDs from Defibtech, Phillips, and Zoll on the determination of ECG rhythm shockability. Performance was evaluated for 181 human ECG recordings (101 ventricular fibrillation-VF, 55 PEA, and 25 asystole) ranging widely in signal characteristics, obtained from the Pittsburgh site of the Resuscitation Outcomes Consortium. We used a commercially available digital-to-analog converter (National Instruments USB-6001) to inject the recordings into each AED through a direct lead-wire interface, recording shock advisement decisions in a best-out-of-three approach for each device/rhythm pairing. We calculated the sensitivity and specificity for discriminating VF and non-VF rhythms for each device and overall. VF signal characteristics were calculated, including peak frequency, median amplitude, and peak amplitude, and the VF quantitative waveform measures AMSA and median slope. Results: The 101 VF trials featured signals with mean peak frequency 10.02 Hz(IQR 4.80 Hz), mean AMSA 9.13(IQR 7.29), mean median slope 6.72 (IQR 3.66). The sensitivities were: Defibtech 99.0%; Philips 97.0%; Zoll 98.0%. The specificities were: Defibtech 98.7%; Philips 96.2%; Zoll 97.4%. Defibtech recorded 5 discordant advisements and Philips and Zoll recorded eight each. The overall sensitivity was 98.0%, and the specificity 97.4%. Conclusion: Evaluated against a wide variety of real-world signal presentations, commercial AEDs demonstrated a high degree of sensitivity and specificity for shockable ECG rhythms.

Deep Learning and Biomedical Engineering

Biotechnology ◽  
2019 ◽  
pp. 562-575
Author(s):  
Suraj Sawant
Keyword(s):  
Machine Learning ◽  
Neural Networks ◽  
Artificial Neural Networks ◽  
Deep Learning ◽  
Real World ◽  
Biomedical Engineering ◽  
Input Data ◽  
Reference Source ◽  
Data Input ◽  
Essential Information

Deep learning (DL) is a method of machine learning, as running over artificial neural networks, which has a structure above the standards to deal with large amounts of data. That is generally because of the increasing amount of data, input data sizes, and of course, greater complexity of objective real-world problems. Performed research studies in the associated literature show that the DL currently has a good performance among considered problems and it seems to be a strong solution for more advanced problems of the future. In this context, this chapter aims to provide some essential information about DL and its applications within the field of biomedical engineering. The chapter is organized as a reference source for enabling readers to have an idea about the relation between DL and biomedical engineering.

Sign in / Sign up

Export Citation Format

Share Document