RESUMEN
PURPOSE OF REVIEW: Technologies in healthcare incorporating artificial intelligence tools are experiencing rapid growth in static-image-based applications such as diagnostic imaging. Given the proliferation of artificial intelligence (AI)-technologies created for video-based imaging, ophthalmic microsurgery is likely to experience significant benefits from the application of emerging technologies to multiple facets of the care of the surgical patient. RECENT FINDINGS: Proof-of-concept research and early phase clinical trials are in progress for AI-based surgical technologies that aim to provide preoperative planning and decision support, intraoperative image enhancement, surgical guidance, surgical decision-making support, tactical assistive technologies, enhanced surgical training and assessment of trainee progress, and semi-autonomous tool control or autonomous elements of surgical procedures. SUMMARY: The proliferation of AI-based technologies in static imaging in clinical ophthalmology, continued refinement of AI tools designed for video-based applications, and development of AI-based digital tools in allied surgical fields suggest that ophthalmic surgery is poised for the integration of AI into our microsurgical paradigm.
RESUMEN
BACKGROUND: Detecting and segmenting brain metastases is a tedious and time-consuming task for many radiologists, particularly with the growing use of multisequence 3D imaging. PURPOSE: To demonstrate automated detection and segmentation of brain metastases on multisequence MRI using a deep-learning approach based on a fully convolution neural network (CNN). STUDY TYPE: Retrospective. POPULATION: In all, 156 patients with brain metastases from several primary cancers were included. FIELD STRENGTH: 1.5T and 3T. [Correction added on May 24, 2019, after first online publication: In the preceding sentence, the first field strength listed was corrected.] SEQUENCE: Pretherapy MR images included pre- and postgadolinium T1 -weighted 3D fast spin echo (CUBE), postgadolinium T1 -weighted 3D axial IR-prepped FSPGR (BRAVO), and 3D CUBE fluid attenuated inversion recovery (FLAIR). ASSESSMENT: The ground truth was established by manual delineation by two experienced neuroradiologists. CNN training/development was performed using 100 and 5 patients, respectively, with a 2.5D network based on a GoogLeNet architecture. The results were evaluated in 51 patients, equally separated into those with few (1-3), multiple (4-10), and many (>10) lesions. STATISTICAL TESTS: Network performance was evaluated using precision, recall, Dice/F1 score, and receiver operating characteristic (ROC) curve statistics. For an optimal probability threshold, detection and segmentation performance was assessed on a per-metastasis basis. The Wilcoxon rank sum test was used to test the differences between patient subgroups. RESULTS: The area under the ROC curve (AUC), averaged across all patients, was 0.98 ± 0.04. The AUC in the subgroups was 0.99 ± 0.01, 0.97 ± 0.05, and 0.97 ± 0.03 for patients having 1-3, 4-10, and >10 metastases, respectively. Using an average optimal probability threshold determined by the development set, precision, recall, and Dice score were 0.79 ± 0.20, 0.53 ± 0.22, and 0.79 ± 0.12, respectively. At the same probability threshold, the network showed an average false-positive rate of 8.3/patient (no lesion-size limit) and 3.4/patient (10 mm3 lesion size limit). DATA CONCLUSION: A deep-learning approach using multisequence MRI can automatically detect and segment brain metastases with high accuracy. LEVEL OF EVIDENCE: 3 Technical Efficacy Stage: 2 J. Magn. Reson. Imaging 2020;51:175-182.
Asunto(s)
Neoplasias Encefálicas/diagnóstico por imagen , Neoplasias Encefálicas/secundario , Aprendizaje Profundo , Interpretación de Imagen Asistida por Computador/métodos , Imagen por Resonancia Magnética/métodos , Adulto , Anciano , Encéfalo/diagnóstico por imagen , Femenino , Humanos , Masculino , Persona de Mediana Edad , Estudios Retrospectivos , Sensibilidad y EspecificidadRESUMEN
The transition from single-cell to multicellular behavior is important in early development but rarely studied. The starvation-induced aggregation of the social amoeba Dictyostelium discoideum into a multicellular slug is known to result from single-cell chemotaxis towards emitted pulses of cyclic adenosine monophosphate (cAMP). However, how exactly do transient, short-range chemical gradients lead to coherent collective movement at a macroscopic scale? Here, we developed a multiscale model verified by quantitative microscopy to describe behaviors ranging widely from chemotaxis and excitability of individual cells to aggregation of thousands of cells. To better understand the mechanism of long-range cell-cell communication and hence aggregation, we analyzed cell-cell correlations, showing evidence of self-organization at the onset of aggregation (as opposed to following a leader cell). Surprisingly, cell collectives, despite their finite size, show features of criticality known from phase transitions in physical systems. By comparing wild-type and mutant cells with impaired aggregation, we found the longest cell-cell communication distance in wild-type cells, suggesting that criticality provides an adaptive advantage and optimally sized aggregates for the dispersal of spores.
Asunto(s)
Quimiotaxis/fisiología , AMP Cíclico/metabolismo , Dictyostelium/metabolismo , Transducción de Señal/fisiología , Algoritmos , Quimiotaxis/genética , Dictyostelium/citología , Espacio Intracelular/metabolismo , Microscopía Fluorescente , Modelos Biológicos , Movimiento/fisiología , Mutación , Transducción de Señal/genética , Imagen de Lapso de Tiempo/métodosRESUMEN
Purpose To assess the ability of convolutional neural networks (CNNs) to enable high-performance automated binary classification of chest radiographs. Materials and Methods In a retrospective study, 216 431 frontal chest radiographs obtained between 1998 and 2012 were procured, along with associated text reports and a prospective label from the attending radiologist. This data set was used to train CNNs to classify chest radiographs as normal or abnormal before evaluation on a held-out set of 533 images hand-labeled by expert radiologists. The effects of development set size, training set size, initialization strategy, and network architecture on end performance were assessed by using standard binary classification metrics; detailed error analysis, including visualization of CNN activations, was also performed. Results Average area under the receiver operating characteristic curve (AUC) was 0.96 for a CNN trained with 200 000 images. This AUC value was greater than that observed when the same model was trained with 2000 images (AUC = 0.84, P < .005) but was not significantly different from that observed when the model was trained with 20 000 images (AUC = 0.95, P > .05). Averaging the CNN output score with the binary prospective label yielded the best-performing classifier, with an AUC of 0.98 (P < .005). Analysis of specific radiographs revealed that the model was heavily influenced by clinically relevant spatial regions but did not reliably generalize beyond thoracic disease. Conclusion CNNs trained with a modestly sized collection of prospectively labeled chest radiographs achieved high diagnostic performance in the classification of chest radiographs as normal or abnormal; this function may be useful for automated prioritization of abnormal chest radiographs. © RSNA, 2018 Online supplemental material is available for this article. See also the editorial by van Ginneken in this issue.
Asunto(s)
Redes Neurales de la Computación , Interpretación de Imagen Radiográfica Asistida por Computador/métodos , Radiografía Torácica/métodos , Femenino , Humanos , Pulmón/diagnóstico por imagen , Masculino , Curva ROC , Radiólogos , Estudios RetrospectivosRESUMEN
Chronic elevation of blood glucose at first causes relatively minor changes to the neural and vascular components of the retina. As the duration of hyperglycemia persists, the nature and extent of damage increases and becomes readily detectable. While this second, overt manifestation of diabetic retinopathy (DR) has been studied extensively, what prevents maximal damage from the very start of hyperglycemia remains largely unexplored. Recent studies indicate that diabetes (DM) engages mitochondria-based defense during the retinopathy-resistant phase, and thereby enables the retina to remain healthy in the face of hyperglycemia. Such resilience is transient, and its deterioration results in progressive accumulation of retinal damage. The concepts that co-emerge with these discoveries set the stage for novel intellectual and therapeutic opportunities within the DR field. Identification of biomarkers and mediators of protection from DM-mediated damage will enable development of resilience-based therapies that will indefinitely delay the onset of DR.
Asunto(s)
Retinopatía Diabética , Humanos , Mitocondrias , Retina , Glucemia/metabolismo , Animales , HiperglucemiaRESUMEN
OBJECTIVE: Uveal melanoma is the most common intraocular malignancy in adults. Current screening and triaging methods for melanocytic choroidal tumours face inherent limitations, particularly in regions with limited access to specialized ocular oncologists. This study explores the potential of machine learning to automate tumour segmentation. We develop and evaluate a machine-learning model for lesion segmentation using ultra-wide-field fundus photography. METHOD: A retrospective chart review was conducted of patients diagnosed with uveal melanoma, choroidal nevi, or congenital hypertrophy of the retinal pigmented epithelium at a tertiary academic medical centre. Included patients had a single ultra-wide-field fundus photograph (Optos PLC, Dunfermline, Fife, Scotland) of adequate quality to visualize the lesion of interest, as confirmed by a single ocular oncologist. These images were used to develop and test a machine-learning algorithm for lesion segmentation. RESULTS: A total of 396 images were used to develop a machine-learning algorithm for lesion segmentation. Ninety additional images were used in the testing data set along with images of 30 healthy control individuals. Of the images with successfully detected lesions, the machine-learning segmentation yielded Dice coefficients of 0.86, 0.81, and 0.85 for uveal melanoma, choroidal nevi, and congenital hypertrophy of the retinal pigmented epithelium, respectively. Sensitivities for any lesion detection per image were 1.00, 0.90, and 0.87, respectively. For images without lesions, specificity was 0.93. CONCLUSION: Our study demonstrates a novel machine-learning algorithm's performance, suggesting its potential clinical utility as a widely accessible method of screening choroidal tumours. Additional evaluation methods are necessary to further enhance the model's lesion classification and diagnostic accuracy.
RESUMEN
Purpose: To develop and validate a platform that can extract eye gaze metrics from surgeons observing cataract and vitreoretinal procedures and to enable post hoc data analysis to assess potential discrepancies in eye movement behavior according to surgeon experience. Design: Experimental, prospective, single-center study. Participants: Eleven ophthalmic surgeons observing deidentified vitreoretinal and cataract surgical procedures performed at a single university-based medical center. Methods: An open-source platform was developed to extract gaze coordinates and metrics from ophthalmic surgeons via a computer vision algorithm in conjunction with a neural network to track and segment instruments and tissues, identifying areas of attention in the visual field of study subjects. Eleven surgeons provided validation data by watching videos of 6 heterogeneous vitreoretinal and cataract surgical phases. Main Outcome Measures: Accuracy and distance traveled by the eye gaze of participants and overlap of the participants' eye gaze with instruments and tissues while observing surgical procedures. Results: The platform demonstrated repeatability of > 94% when acquiring the eye gaze behavior of subjects. Attending ophthalmic surgeons and clinical fellows exhibited a lower overall cartesian distance traveled in comparison to resident physicians in ophthalmology (P < 0.02). Ophthalmology residents and clinical fellows exhibited more fixations to the display area where surgical device parameters were superimposed than attending surgeons (P < 0.05). There was a trend toward gaze overlap with the instrument tooltip location among resident physicians in comparison to attending surgeons and fellows (41.42% vs. 34.8%, P > 0.2). The number and duration of fixations did not vary substantially among groups (P > 0.3). Conclusions: The platform proved effective in extracting gaze metrics of ophthalmic surgeons. These preliminary data suggest that surgeon gaze behavior differs according to experience.
RESUMEN
PURPOSE: This study investigated whether a deep-learning neural network can detect and segment surgical instrumentation and relevant tissue boundaries and landmarks within the retina using imaging acquired from a surgical microscope in real time, with the goal of providing image-guided vitreoretinal (VR) microsurgery. DESIGN: Retrospective analysis via a prospective, single-center study. PARTICIPANTS: One hundred and one patients undergoing VR surgery, inclusive of core vitrectomy, membrane peeling, and endolaser application, in a university-based ophthalmology department between July 1, 2020, and September 1, 2021. METHODS: A dataset composed of 606 surgical image frames was annotated by 3 VR surgeons. Annotation consisted of identifying the location and area of the following features, when present in-frame: vitrector-, forceps-, and endolaser tooltips, optic disc, fovea, retinal tears, retinal detachment, fibrovascular proliferation, endolaser spots, area where endolaser was applied, and macular hole. An instance segmentation fully convolutional neural network (YOLACT++) was adapted and trained, and fivefold cross-validation was employed to generate metrics for accuracy. MAIN OUTCOME MEASURES: Area under the precision-recall curve (AUPR) for the detection of elements tracked and segmented in the final test dataset; the frames per second (FPS) for the assessment of suitability for real-time performance of the model. RESULTS: The platform detected and classified the vitrector tooltip with a mean AUPR of 0.972 ± 0.009. The segmentation of target tissues, such as the optic disc, fovea, and macular hole reached mean AUPR values of 0.928 ± 0.013, 0.844 ± 0.039, and 0.916 ± 0.021, respectively. The postprocessed image was rendered at a full high-definition resolution of 1920 × 1080 pixels at 38.77 ± 1.52 FPS when attached to a surgical visualization system, reaching up to 87.44 ± 3.8 FPS. CONCLUSIONS: Neural Networks can localize, classify, and segment tissues and instruments during VR procedures in real time. We propose a framework for developing surgical guidance and assessment platform that may guide surgical decision-making and help in formulating tools for systematic analyses of VR surgery. Potential applications include collision avoidance to prevent unintended instrument-tissue interactions and the extraction of spatial localization and movement of surgical instruments for surgical data science research. FINANCIAL DISCLOSURE(S): Proprietary or commercial disclosure may be found after the references.
Asunto(s)
Aprendizaje Profundo , Oftalmología , Perforaciones de la Retina , Cirugía Vitreorretiniana , Humanos , Inteligencia Artificial , Estudios Retrospectivos , Estudios ProspectivosRESUMEN
Purpose: This study aims to investigate generalizability of deep learning (DL) models trained on commonly used public fundus images to an instance of real-world data (RWD) for glaucoma diagnosis. Methods: We used Illinois Eye and Ear Infirmary fundus data set as an instance of RWD in addition to six publicly available fundus data sets. We compared the performance of DL-trained models on public data and RWD for glaucoma classification and optic disc (OD) segmentation tasks. For each task, we created models trained on each data set, respectively, and each model was tested on both data sets. We further examined each model's decision-making process and learned embeddings for the glaucoma classification task. Results: Using public data for the test set, public-trained models outperformed RWD-trained models in OD segmentation and glaucoma classification with a mean intersection over union of 96.3% and mean area under the receiver operating characteristic curve of 95.0%, respectively. Using the RWD test set, the performance of public models decreased by 8.0% and 18.4% to 85.6% and 76.6% for OD segmentation and glaucoma classification tasks, respectively. RWD models outperformed public models on RWD test sets by 2.0% and 9.5%, respectively, in OD segmentation and glaucoma classification tasks. Conclusions: DL models trained on commonly used public data have limited ability to generalize to RWD for classifying glaucoma. They perform similarly to RWD models for OD segmentation. Translational Relevance: RWD is a potential solution for improving generalizability of DL models and enabling clinical translations in the care of prevalent blinding ophthalmic conditions, such as glaucoma.
Asunto(s)
Aprendizaje Profundo , Glaucoma , Disco Óptico , Humanos , Inteligencia Artificial , Disco Óptico/diagnóstico por imagen , Glaucoma/diagnóstico , Fondo de OjoRESUMEN
Objective: This study aims to assess a machine learning (ML) algorithm using multimodal imaging to accurately identify risk factors for uveal melanoma (UM) and aid in the diagnosis of melanocytic choroidal tumors. Subjects and Methods: This study included 223 eyes from 221 patients with melanocytic choroidal lesions seen at the eye clinic of the University of Illinois at Chicago between 01/2010 and 07/2022. An ML algorithm was developed and trained on ultra-widefield fundus imaging and B-scan ultrasonography to detect risk factors of malignant transformation of choroidal lesions into UM. The risk factors were verified using all multimodal imaging available from the time of diagnosis. We also explore classification of lesions into UM and choroidal nevi using the ML algorithm. Results: The ML algorithm assessed features of ultra-widefield fundus imaging and B-scan ultrasonography to determine the presence of the following risk factors for malignant transformation: lesion thickness, subretinal fluid, orange pigment, proximity to optic nerve, ultrasound hollowness, and drusen. The algorithm also provided classification of lesions into UM and choroidal nevi. A total of 115 patients with choroidal nevi and 108 patients with UM were included. The mean lesion thickness for choroidal nevi was 1.6 mm and for UM was 5.9 mm. Eleven ML models were implemented and achieved high accuracy, with an area under the curve of 0.982 for thickness prediction and 0.964 for subretinal fluid prediction. Sensitivity/specificity values ranged from 0.900/0.818 to 1.000/0.727 for different features. The ML algorithm demonstrated high accuracy in identifying risk factors and differentiating lesions based on the analyzed imaging data. Conclusions: This study provides proof of concept that ML can accurately identify risk factors for malignant transformation in melanocytic choroidal tumors based on a single ultra-widefield fundus image or B-scan ultrasound at the time of initial presentation. By leveraging the efficiency and availability of ML, this study has the potential to provide a non-invasive tool that helps to prevent unnecessary treatment, improve our ability to predict malignant transformation, reduce the risk of metastasis, and potentially save patient lives.
RESUMEN
IMPORTANCE: Complications that arise from phacoemulsification procedures can lead to worse visual outcomes. Real-time image processing with artificial intelligence tools can extract data to deliver surgical guidance, potentially enhancing the surgical environment. OBJECTIVE: To evaluate the ability of a deep neural network to track the pupil, identify the surgical phase, and activate specific computer vision tools to aid the surgeon during phacoemulsification cataract surgery by providing visual feedback in real time. DESIGN, SETTING, AND PARTICIPANTS: This cross-sectional study evaluated deidentified surgical videos of phacoemulsification cataract operations performed by faculty and trainee surgeons in a university-based ophthalmology department between July 1, 2020, and January 1, 2021, in a population-based cohort of patients. EXPOSURES: A region-based convolutional neural network was used to receive frames from the video source and, in real time, locate the pupil and in parallel identify the surgical phase being performed. Computer vision-based algorithms were applied according to the phase identified, providing visual feedback to the surgeon. MAIN OUTCOMES AND MEASURES: Outcomes were area under the receiver operator characteristic curve and area under the precision-recall curve for surgical phase classification and Dice score (harmonic mean of the precision and recall [sensitivity]) for detection of the pupil boundary. Network performance was assessed as video output in frames per second. A usability survey was administered to volunteer cataract surgeons previously unfamiliar with the platform. RESULTS: The region-based convolutional neural network model achieved area under the receiver operating characteristic curve values of 0.996 for capsulorhexis, 0.972 for phacoemulsification, 0.997 for cortex removal, and 0.880 for idle phase recognition. The final algorithm reached a Dice score of 90.23% for pupil segmentation and a mean (SD) processing speed of 97 (34) frames per second. Among the 11 cataract surgeons surveyed, 8 (72%) were mostly or extremely likely to use the current platform during surgery for complex cataract. CONCLUSIONS AND RELEVANCE: A computer vision approach using deep neural networks was able to pupil track, identify the surgical phase being executed, and activate surgical guidance tools. These results suggest that an artificial intelligence-based surgical guidance platform has the potential to enhance the surgeon experience in phacoemulsification cataract surgery. This proof-of-concept investigation suggests that a pipeline from a surgical microscope could be integrated with neural networks and computer vision tools to provide surgical guidance in real time.
Asunto(s)
Catarata , Oftalmología , Facoemulsificación , Inteligencia Artificial , Estudios Transversales , Humanos , Facoemulsificación/métodosRESUMEN
Introduction: Management of patients with brain metastases is often based on manual lesion detection and segmentation by an expert reader. This is a time- and labor-intensive process, and to that end, this work proposes an end-to-end deep learning segmentation network for a varying number of available MRI available sequences. Methods: We adapt and evaluate a 2.5D and a 3D convolution neural network trained and tested on a retrospective multinational study from two independent centers, in addition, nnU-Net was adapted as a comparative benchmark. Segmentation and detection performance was evaluated by: (1) the dice similarity coefficient, (2) a per-metastases and the average detection sensitivity, and (3) the number of false positives. Results: The 2.5D and 3D models achieved similar results, albeit the 2.5D model had better detection rate, whereas the 3D model had fewer false positive predictions, and nnU-Net had fewest false positives, but with the lowest detection rate. On MRI data from center 1, the 2.5D, 3D, and nnU-Net detected 79%, 71%, and 65% of all metastases; had an average per patient sensitivity of 0.88, 0.84, and 0.76; and had on average 6.2, 3.2, and 1.7 false positive predictions per patient, respectively. For center 2, the 2.5D, 3D, and nnU-Net detected 88%, 86%, and 78% of all metastases; had an average per patient sensitivity of 0.92, 0.91, and 0.85; and had on average 1.0, 0.4, and 0.1 false positive predictions per patient, respectively. Discussion/Conclusion: Our results show that deep learning can yield highly accurate segmentations of brain metastases with few false positives in multinational data, but the accuracy degrades for metastases with an area smaller than 0.4 cm2.
RESUMEN
PURPOSE: Magnetic resonance (MR) imaging is an essential diagnostic tool in clinical medicine. Recently, a variety of deep-learning methods have been applied to segmentation tasks in medical images, with promising results for computer-aided diagnosis. For MR images, effectively integrating different pulse sequences is important to optimize performance. However, the best way to integrate different pulse sequences remains unclear. In addition, networks trained with a certain subset of pulse sequences as input are unable to perform when given a subset of those pulse sequences. In this study, we evaluate multiple architectural features and characterize their effects in the task of metastasis segmentation while creating a method to robustly train a network to be able to work given any strict subset of the pulse sequences available during training. METHODS: We use a 2.5D DeepLabv3 segmentation network to segment metastases lesions on brain MR's with four pulse sequence inputs. To study how we can best integrate MR pulse sequences for this task, we consider (1) different pulse sequence integration schemas, combining our features at early, middle, and late points within a deep network, (2) different modes of weight sharing for parallel network branches, and (3) a novel integration level dropout layer, which will allow the networks to be robust to performing inference on input with only a subset of pulse sequences available at the training. RESULTS: We find that levels of integration and modes of weight sharing that favor low variance work best in our regime of small amounts of training data (n = 100). By adding an input-level dropout layer, we could preserve the overall performance of these networks while allowing for inference on inputs with missing pulse sequences. We illustrate not only the generalizability of the network but also the utility of this robustness when applying the trained model to data from a different center, which does not use the same pulse sequences. Finally, we apply network visualization methods to better understand which input features are most important for network performance. CONCLUSIONS: Together, these results provide a framework for building networks with enhanced robustness to missing data while maintaining comparable performance in medical imaging applications.
Asunto(s)
Neoplasias Encefálicas , Aprendizaje Profundo , Neoplasias Encefálicas/diagnóstico por imagen , Humanos , Procesamiento de Imagen Asistido por Computador , Imagen por Resonancia Magnética , Redes Neurales de la ComputaciónRESUMEN
The purpose of this study was to assess the clinical value of a deep learning (DL) model for automatic detection and segmentation of brain metastases, in which a neural network is trained on four distinct MRI sequences using an input-level dropout layer, thus simulating the scenario of missing MRI sequences by training on the full set and all possible subsets of the input data. This retrospective, multicenter study, evaluated 165 patients with brain metastases. The proposed input-level dropout (ILD) model was trained on multisequence MRI from 100 patients and validated/tested on 10/55 patients, in which the test set was missing one of the four MRI sequences used for training. The segmentation results were compared with the performance of a state-of-the-art DeepLab V3 model. The MR sequences in the training set included pre-gadolinium and post-gadolinium (Gd) T1-weighted 3D fast spin echo, post-Gd T1-weighted inversion recovery (IR) prepped fast spoiled gradient echo, and 3D fluid attenuated inversion recovery (FLAIR), whereas the test set did not include the IR prepped image-series. The ground truth segmentations were established by experienced neuroradiologists. The results were evaluated using precision, recall, Intersection over union (IoU)-score and Dice score, and receiver operating characteristics (ROC) curve statistics, while the Wilcoxon rank sum test was used to compare the performance of the two neural networks. The area under the ROC curve (AUC), averaged across all test cases, was 0.989 ± 0.029 for the ILD-model and 0.989 ± 0.023 for the DeepLab V3 model (p = 0.62). The ILD-model showed a significantly higher Dice score (0.795 ± 0.104 vs. 0.774 ± 0.104, p = 0.017), and IoU-score (0.561 ± 0.225 vs. 0.492 ± 0.186, p < 0.001) compared to the DeepLab V3 model, and a significantly lower average false positive rate of 3.6/patient vs. 7.0/patient (p < 0.001) using a 10 mm3 lesion-size limit. The ILD-model, trained on all possible combinations of four MRI sequences, may facilitate accurate detection and segmentation of brain metastases on a multicenter basis, even when the test cohort is missing input MRI sequences.
RESUMEN
Despite the relative ease of locating organs in the human body, automated organ segmentation has been hindered by the scarcity of labeled training data. Due to the tedium of labeling organ boundaries, most datasets are limited to either a small number of cases or a single organ. Furthermore, many are restricted to specific imaging conditions unrepresentative of clinical practice. To address this need, we developed a diverse dataset of 140 CT scans containing six organ classes: liver, lungs, bladder, kidney, bones and brain. For the lungs and bones, we expedited annotation using unsupervised morphological segmentation algorithms, which were accelerated by 3D Fourier transforms. Demonstrating the utility of the data, we trained a deep neural network which requires only 4.3 s to simultaneously segment all the organs in a case. We also show how to efficiently augment the data to improve model generalization, providing a GPU library for doing so. We hope this dataset and code, available through TCIA, will be useful for training and evaluating organ segmentation models.
Asunto(s)
Procesamiento de Imagen Asistido por Computador/métodos , Redes Neurales de la Computación , Tomografía Computarizada por Rayos X , Algoritmos , Huesos/diagnóstico por imagen , Encéfalo/diagnóstico por imagen , Análisis de Fourier , Humanos , Imagenología Tridimensional/métodos , Riñón/diagnóstico por imagen , Hígado/diagnóstico por imagen , Pulmón/diagnóstico por imagen , Vejiga Urinaria/diagnóstico por imagenRESUMEN
Importance: Mammography screening currently relies on subjective human interpretation. Artificial intelligence (AI) advances could be used to increase mammography screening accuracy by reducing missed cancers and false positives. Objective: To evaluate whether AI can overcome human mammography interpretation limitations with a rigorous, unbiased evaluation of machine learning algorithms. Design, Setting, and Participants: In this diagnostic accuracy study conducted between September 2016 and November 2017, an international, crowdsourced challenge was hosted to foster AI algorithm development focused on interpreting screening mammography. More than 1100 participants comprising 126 teams from 44 countries participated. Analysis began November 18, 2016. Main Outcomes and Measurements: Algorithms used images alone (challenge 1) or combined images, previous examinations (if available), and clinical and demographic risk factor data (challenge 2) and output a score that translated to cancer yes/no within 12 months. Algorithm accuracy for breast cancer detection was evaluated using area under the curve and algorithm specificity compared with radiologists' specificity with radiologists' sensitivity set at 85.9% (United States) and 83.9% (Sweden). An ensemble method aggregating top-performing AI algorithms and radiologists' recall assessment was developed and evaluated. Results: Overall, 144â¯231 screening mammograms from 85â¯580 US women (952 cancer positive ≤12 months from screening) were used for algorithm training and validation. A second independent validation cohort included 166â¯578 examinations from 68â¯008 Swedish women (780 cancer positive). The top-performing algorithm achieved an area under the curve of 0.858 (United States) and 0.903 (Sweden) and 66.2% (United States) and 81.2% (Sweden) specificity at the radiologists' sensitivity, lower than community-practice radiologists' specificity of 90.5% (United States) and 98.5% (Sweden). Combining top-performing algorithms and US radiologist assessments resulted in a higher area under the curve of 0.942 and achieved a significantly improved specificity (92.0%) at the same sensitivity. Conclusions and Relevance: While no single AI algorithm outperformed radiologists, an ensemble of AI algorithms combined with radiologist assessment in a single-reader screening environment improved overall accuracy. This study underscores the potential of using machine learning methods for enhancing mammography screening interpretation.
Asunto(s)
Neoplasias de la Mama/diagnóstico por imagen , Aprendizaje Profundo , Interpretación de Imagen Asistida por Computador/métodos , Mamografía/métodos , Radiólogos , Adulto , Anciano , Algoritmos , Inteligencia Artificial , Detección Precoz del Cáncer , Femenino , Humanos , Persona de Mediana Edad , Radiología , Sensibilidad y Especificidad , Suecia , Estados UnidosRESUMEN
Diabetic retinopathy is a leading cause of blindness among working-age adults. Early detection of this condition is critical for good prognosis. In this paper, we demonstrate the use of convolutional neural networks (CNNs) on color fundus images for the recognition task of diabetic retinopathy staging. Our network models achieved test metric performance comparable to baseline literature results, with validation sensitivity of 95%. We additionally explored multinomial classification models, and demonstrate that errors primarily occur in the misclassification of mild disease as normal due to the CNNs inability to detect subtle disease features. We discovered that preprocessing with contrast limited adaptive histogram equalization and ensuring dataset fidelity by expert verification of class labels improves recognition of subtle features. Transfer learning on pretrained GoogLeNet and AlexNet models from ImageNet improved peak test set accuracies to 74.5%, 68.8%, and 57.2% on 2-ary, 3-ary, and 4-ary classification models, respectively.
RESUMEN
Objective: Deep learning has become a promising approach for automated support for clinical diagnosis. When medical data samples are limited, collaboration among multiple institutions is necessary to achieve high algorithm performance. However, sharing patient data often has limitations due to technical, legal, or ethical concerns. In this study, we propose methods of distributing deep learning models as an attractive alternative to sharing patient data. Methods: We simulate the distribution of deep learning models across 4 institutions using various training heuristics and compare the results with a deep learning model trained on centrally hosted patient data. The training heuristics investigated include ensembling single institution models, single weight transfer, and cyclical weight transfer. We evaluated these approaches for image classification in 3 independent image collections (retinal fundus photos, mammography, and ImageNet). Results: We find that cyclical weight transfer resulted in a performance that was comparable to that of centrally hosted patient data. We also found that there is an improvement in the performance of cyclical weight transfer heuristic with a high frequency of weight transfer. Conclusions: We show that distributing deep learning models is an effective alternative to sharing patient data. This finding has implications for any collaborative deep learning study.
Asunto(s)
Aprendizaje Profundo , Diagnóstico por Imagen , Redes de Comunicación de Computadores , Humanos , Registro Médico Coordinado , Redes Neurales de la ComputaciónRESUMEN
Data augmentation is an essential part of training discriminative Convolutional Neural Networks (CNNs). A variety of augmentation strategies, including horizontal flips, random crops, and principal component analysis (PCA), have been proposed and shown to capture important characteristics of natural images. However, while data augmentation has been commonly used for deep learning in medical imaging, little work has been done to determine which augmentation strategies best capture medical image statistics, leading to more discriminative models. This work compares augmentation strategies and shows that the extent to which an augmented training set retains properties of the original medical images determines model performance. Specifically, augmentation strategies such as flips and gaussian filters lead to validation accuracies of 84% and 88%, respectively. On the other hand, a less effective strategy such as adding noise leads to a significantly worse validation accuracy of 66%. Finally, we show that the augmentation affects mass generation.