Fully automated mouse echocardiography analysis using deep convolutional neural networks

Abstract

Echocardiography (echo) is a translationally relevant ultrasound imaging modality widely used to assess cardiac structure and function in preclinical models of heart failure (HF) during research and drug development. Though echo is a very valuable tool, the image analysis is a time consuming, resource demanding process, and is susceptible to inter-reader variability. Recent advancements in deep learning have enabled researchers to automate image processing and reduce analysis time and inter-reader variability in the field of medical imaging. In the present study, we developed a fully automated tool - Mouse Echo Neural Net (MENN) - for the analysis of both long axis brightness (B)-mode and short axis motion (M)-mode images of the left ventricle. MENN is a series of fully convolutional neural networks that were trained and validated using manually segmented B-mode and M-mode echo images of the left ventricle. The segmented images were then used to compute cardiac structural and functional metrics. The performance of MENN was further validated in two preclinical models of HF. MENN achieved excellent correlations (Pearson's r = 0.85 to 0.99) and good to excellent agreement between automated and manual analyses. Further inter-reader variability analysis showed that MENN has better agreements with an expert analyst than both a trained analyst and a novice. Notably, the use of MENN reduced manual analysis time by >92%. In conclusion, we developed an automated echocardiography analysis tool that allows for fast and accurate analysis of B-mode and M-mode mouse echo data and mitigates the issue of inter-reader variability in manual analysis.

Applications of Deep Learning Algorithms to Ultrasound Imaging Analysis in Preclinical Studies on In Vivo Animals

“…Following the review of the title and abstract and, upon necessity, the full text further 187 records were rejected. After the reviewing process, a total of 36 original papers [ 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 , 54 , 55 , 56 , 57 , 58 , 59 , 60 , 61 , 62 , 63 ] and 20 conference proceedings [ 64 , 65 , 66 , 67 , 68 , 69 , 70 , 71 , 72 , 73 , 74 , 75 , 76 , 77 , 78 , 79 , 80 , 81 , 82 ,…”

Applying deep learning to segmentation of murine lung tumors in pre-clinical micro-computed tomography

“…and Duan et al. for the purposes of this model [ 13 , 39 ]. The model had four encoding layers and 4 decoding layers each consisting of two convolutional layers with a 3 × 3 kernel size and a max-pooling or up-convolution layer, respectively.…”

AData‐DrivenAnalysis of Myocardial Tissue Contraction on Mild Transverse Aortic Constriction

“…In some M‐mode images, papillary muscles are located on the boundaries of the ventricle, making segmentation based on pixel information alone difficult [10]. When papillary muscle appeared in the imaging view, conventional segmentation methods such as the snakes method [11] could not be applied.…”