Temporal stability assessment in shear wave elasticity images validated by deep learning neural network for chronic liver disease fibrosis stage assessment

• DOI: 10.1002/mp.13521
• Publication Date: 2019-03-31T10:43:00Z
• Authors: Ilias Gatos; Stavros Tsantis; Stavros Spiliopoulos; Dimitris Karnabat...; Ioannis Theotokas; Pavlos Zoumpoulis; Thanasis Loupas; John D. Hazle; George C. Kagadis
• Keywords: Chronic liver disease

Purpose To automatically detect and isolate areas of low high stiffness temporal stability in shear wave elastography ( SWE ) image sequences define their impact chronic liver disease CLD diagnosis improvement by means clinical examination study deep learning algorithm employing convolutional neural networks CNN s). Materials Methods Two hundred from 88 healthy individuals (F0 fibrosis stage) 112 patients (46 with mild (F1), 16 significant (F2), 22 severe (F3), 28 cirrhosis (F4)) were analyzed to between frames. An inverse Red, Green, Blue RGB colormap‐to‐stiffness process was performed for each sequence, followed a wavelet transform fuzzy c‐means clustering algorithm. This resulted binary mask depicting stability. The then applied the first derived, masked used estimate its standard classification. Regarding examination, one measurement two radiologists corresponding measuring Then, parameters, interobserver variability evaluation diagnostic performance ROC analysis assessed. unmasked sets images fed into scheme comparison. Results showed that decreased radiologists’ measurements (interclass correlation coefficient ICC = 0.92) compared ones 0.76). In terms accuracy, high‐stability (area‐under‐the‐curve AUC ranging 0.800 0.851) similarly those 0.805 0.893). images, results 0.63) accuracy 0.622 0.791) poor. classification, improved (ranging 82.5% 95.5%) 79.5% 93.2%) various stage combinations. Conclusion Our detection excludes unreliable reduces variability, augments 's scores many combinations stages.