Deep Learning for Accurate Diagnosis of Liver Tumor Based on Magnetic Resonance Imaging and Clinical Data
Глубокое обучение для точной диагностики опухолей печени на основе магнитно-резонансной томографии и клинических данных
2020-05-28
SCID: 54.1/73thcyq8
Discuss with AI
clinical dataconvolutional neural networksdeep learningliver tumor classificationmagnetic resonance imaging
Figures from the paper
Abstract (AI)
Background: Early-stage diagnosis and treatment can improve survivals of liver cancer patients. Dynamic contrast-enhanced MRI provides the most comprehensive information for differential diagnosis of liver tumors. However, MRI diagnosis is affected by subjective experience, deep learning may supply a new diagnostic strategy. We used convolutional neural networks(CNNs) to develop a deep learning system(DLS) to classify liver tumors basing-on enhanced MR images, unenhanced MR images, and clinical data including text and laboratory test results.2 This is a provisional file, not the final typeset article Methods: Using data from 1210 patients with liver tumors(N = 31608 images), we trained CNNs to get seven-way classifiers, binary classifiers and three-way malignancy-classifiers (Model A-Model G). Models were validated in an external independent extended cohort of 201 patients( N = 6816 images). Area under receiver operating characteristic (ROC) curve (AUC) were compared across different models. We also compared the sensitivity and specificity of models with the performance of three experienced radiologists.Results: Deep learning achieves performance on par with three experienced radiologists on classifying liver tumors to seven categories. Using only un-enhanced images, CNN performs well to distinguish malignant from benign liver tumors (AUC of 0·946(95% CI 0·914-0·979) vs 0·951 (0·919-0·982), P=0.664). New CNN combining unenhanced images with clinical data greatly improved the performance of classifying malignancy as hepatocellular carcinoma(AUC of 0.985 (0.960-1.000)),metastatic tumors(0.998 (0.989-1.000)) and other primary malignancy(0.963 (0.896-1.000)), and the agreement with pathology was 91.9%.These models mined diagnostic information in unenhanced images and clinical data by deep-neural-network, which were different with previous method that utilized enhanced images. The sensitivity and specificity of almost every category in these models reached the same high level compared to three experienced radiologists.Trained with data in various acquisition condition, DLS that integrated these models could be served as an accurate and time-saving assisted-diagnostic strategy for liver tumors in clinical setting even in absence of contrast agents. DLS therefore has the potential to avoid contrast-related side effects and reduce economic costs associated with current standard MRI inspection practices for liver tumor patients.
Key Findings
1
A convolutional-neural-network system classified liver tumors into seven categories with performance comparable to three experienced radiologists.
2
Combining unenhanced MRI with clinical data substantially improved malignancy classification, achieving AUCs of 0.985 for hepatocellular carcinoma, 0.998 for metastatic tumors, and 0.963 for other primary malignancies.
3
The combined models showed 91.9% agreement with pathology and extracted diagnostic information from unenhanced images and clinical data rather than relying primarily on contrast-enhanced imaging.
4
The system was developed using 1,210 patients and 31,608 images and validated in an independent cohort of 201 patients with 6,816 images.
5
Using only unenhanced MRI, the CNN distinguished malignant from benign liver tumors with an AUC of 0.946 (95% CI 0.914–0.979), comparable to enhanced-imaging performance (AUC 0.951; P=0.664).
Research Object
Liver tumors characterized by magnetic resonance images and clinical data
Research Subject
Deep-learning classification and malignancy discrimination of liver tumors, including seven-category tumor diagnosis and identification of hepatocellular carcinoma, metastatic tumors, and other primary malignancies
Publication Details
Publication Date
2020-05-28
Journal
Publisher
ISSN
Cited by
226
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest