Dataset Information

Deep learning in rare disease. Detection of tubers in tuberous sclerosis complex.

ABSTRACT: OBJECTIVE:To develop and test a deep learning algorithm to automatically detect cortical tubers in magnetic resonance imaging (MRI), to explore the utility of deep learning in rare disorders with limited data, and to generate an open-access deep learning standalone application. METHODS:T2 and FLAIR axial images with and without tubers were extracted from MRIs of patients with tuberous sclerosis complex (TSC) and controls, respectively. We trained three different convolutional neural network (CNN) architectures on a training dataset and selected the one with the lowest binary cross-entropy loss in the validation dataset, which was evaluated on the testing dataset. We visualized image regions most relevant for classification with gradient-weighted class activation maps (Grad-CAM) and saliency maps. RESULTS:114 patients with TSC and 114 controls were divided into a training set, a validation set, and a testing set. The InceptionV3 CNN architecture performed best in the validation set and was evaluated in the testing set with the following results: sensitivity: 0.95, specificity: 0.95, positive predictive value: 0.94, negative predictive value: 0.95, F1-score: 0.95, accuracy: 0.95, and area under the curve: 0.99. Grad-CAM and saliency maps showed that tubers resided in regions most relevant for image classification within each image. A stand-alone trained deep learning App was able to classify images using local computers with various operating systems. CONCLUSION:This study shows that deep learning algorithms are able to detect tubers in selected MRI images, and deep learning can be prudently applied clinically to manually selected data in a rare neurological disorder.

SUBMITTER: Sanchez Fernandez I

PROVIDER: S-EPMC7190137 | biostudies-literature | 2020

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Deep learning in rare disease. Detection of tubers in tuberous sclerosis complex.

Sánchez Fernández Iván I Yang Edward E Calvachi Paola P Amengual-Gual Marta M Wu Joyce Y JY Krueger Darcy D Northrup Hope H Bebin Martina E ME Sahin Mustafa M Yu Kun-Hsing KH Peters Jurriaan M JM

PloS one 20200429 4

<h4>Objective</h4>To develop and test a deep learning algorithm to automatically detect cortical tubers in magnetic resonance imaging (MRI), to explore the utility of deep learning in rare disorders with limited data, and to generate an open-access deep learning standalone application.<h4>Methods</h4>T2 and FLAIR axial images with and without tubers were extracted from MRIs of patients with tuberous sclerosis complex (TSC) and controls, respectively. We trained three different convolutional neur ...[more]

PMID: 32348367

Similar Datasets

Project description:BackgroundTuberous sclerosis complex (TSC) is a rare multisystem autosomal dominant disorder caused by pathogenic variants in either the TSC1 or TSC2 gene. Common manifestations of TSC have been grouped into major and minor clinical diagnostic criteria and assessed in clinical routine workup. However, case studies point towards the existence of rare disease manifestations and to the potential association of TSC with malignant tumors. In this study we sought to characterize rare manifestations and malignancies using a large cohort of patients.MethodsTuberOus SClerosis registry to increAse disease awareness (TOSCA) is a multicenter, international disease registry collecting clinical manifestations and characteristics of patients with TSC, both retrospectively and prospectively. We report rates and characteristics of rare manifestations and malignancies in patients with TSC who had enrolled in the TOSCA registry. We also examined these manifestations by age, sex, and genotype (TSC1 or TSC2).ResultsOverall, 2211 patients with TSC were enrolled in the study. Rare manifestations were reported in 382 (17.3%) study participants and malignancies in 65 (2.9%). Of these rare manifestations, the most frequent were bone sclerotic foci (39.5%), scoliosis (23%), thyroid adenoma (5.5%), adrenal angiomyolipoma (4.5%), hemihypertrophy and pancreatic neuroendocrine tumors (pNET; both 3.1%). These rare manifestations were more commonly observed in adults than children (66.2% vs. 22.7%), in females versus males (58.4% vs. 41.6%; except for scoliosis: 48.9% vs. 51.1%), and in those with TSC2 versus TSC1 (67.0% vs. 21.1%; except for thyroid adenoma: 42.9% vs. 57.1%). In the 65 individuals with reported malignancies, the most common were renal cell carcinoma (47.7%), followed by breast (10.8%) and thyroid cancer (9.2%). Although malignancies were more common in adult patients, 26.1% were reported in children and 63.1% in individuals < 40 years. TSC1 mutations were over-represented in individuals with malignancies compared to the overall TOSCA cohort (32.1% vs. 18.5%).ConclusionRare manifestations were observed in a significant proportion of individuals with TSC. We recommend further examination of rare manifestations in TSC. Collectively, malignancies were infrequent findings in our cohort. However, compared to the general population, malignant tumors occurred earlier in age and some tumor types were more common.

Project description:BACKGROUND:Renal angiomyolipoma occurs at a high frequency in patients with tuberous sclerosis complex (TSC) and is associated with potentially life-threatening complications. Despite this frequency and severity, there are no large population-based cohort studies. Here we present baseline and follow-up data of the international TuberOus SClerosis registry to increase disease Awareness (TOSCA) with an aim to provide detailed clinical characteristics of renal angiomyolipoma among patients with TSC. METHODS:Patients of any age with a documented clinic visit for TSC within 12 months or who were newly diagnosed with TSC before participation in the registry were eligible. Data specific to renal angiomyolipoma included physical tumour characteristics (multiple, bilateral, lesion size and growing lesions), clinical signs and symptoms, and management. The effects of age, gender and genotype on the prevalence of renal angiomyolipoma were also evaluated. RESULTS:Renal angiomyolipoma was reported in 51.8% of patients at baseline, with higher frequency in female patients (57.8% versus 42.2%). The median age at diagnosis was 12 years. Prevalence of angiomyolipoma was higher in patients with TSC2 compared with TSC1 mutations (59.2% versus 33.3%, P < 0.01). Of the 1031 patients with angiomyolipoma at baseline, multiple lesions were reported in 88.4% and bilateral in 83.9% of patients, while the size of angiomyolipoma was >3 cm in 34.3% of patients. Most patients were asymptomatic (82%). Frequently reported angiomyolipoma-related symptoms included bleeding, pain, elevated blood pressure and impaired renal function. Embolization and mammalian target of rapamycin inhibitors were the two most common treatment modalities. CONCLUSIONS:The TOSCA registry highlights the burden of renal angiomyolipoma in patients with TSC and shows that renal manifestations are initially asymptomatic and are influenced by gender and genotype. Furthermore, the occurrence of significant problems from angiomyolipoma in a minority of younger patients suggests that surveillance should begin in infancy or at initial diagnosis.

Dataset Information

Deep learning in rare disease. Detection of tubers in tuberous sclerosis complex.

Publications

Deep learning in rare disease. Detection of tubers in tuberous sclerosis complex.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets