Automatic muscle perimysium annotation using deep convolutional neural network

Abstract

Diseased skeletal muscle expresses mononuclear cell infiltration in the regions of perimysium. Accurate annotation or segmentation of perimysium can help biologists and clinicians to determine individualized patient treatment and allow for reasonable prognostication. However, manual perimysium annotation is time consuming and prone to inter-observer variations. Meanwhile, the presence of ambiguous patterns in muscle images significantly challenge many traditional automatic annotation algorithms. In this paper, we propose an automatic perimysium annotation algorithm based on deep convolutional neural network (CNN). We formulate the automatic annotation of perimysium in muscle images as a pixel-wise classification problem, and the CNN is trained to label each image pixel with raw RGB values of the patch centered at the pixel. The algorithm is applied to 82 diseased skeletal muscle images. We have achieved an average precision of 94% on the test dataset.

AIIMDs: An Integrated Framework of Automatic Idiopathic Inflammatory Myopathy Diagnosis for Muscle

“…We compare the performance of the proposed SCW-RNN with several other state-of-the-arts deep learning based methods, including: CNN-NMS [45], which is a famous 10-layer architecture used to segment neuronal membranes and uses a large input window (95 × 95) to capture the context information; and our previous work, CNNP [20], which is an end-to-end CNN architecture. To further demonstrate the capability of our proposed method to handle the spatial context information, we train a plain MLP network (MLP-10) that shares a similar architecture for comparison.…”

“…Due to the high variability of the patterns shown in histopathology images, it is difficult to design robust feature descriptors for automatic skeletal muscle image analysis. On the other hand, there is an encouraging evidence that automatically learned representation of biomedical images using deep neural network usually outperforms the handcrafted features in a wide range of applications such as detection, segmentation and diagnosis of different diseases [20]. However, the sliding window-based approaches [20] fail in modeling the global semantic information by exploiting the context information, which could improve performance of perimysium annotation.…”

“…Row 1, from left to right: The original image, and results obtained using MLP-48 and CNNP [20]; Row 2, from left to right: The ground-truth annotation provided by human experts, and results obtained using CNN-NMS [45] and our proposed SCW-RNN(8), respectively.…”

Comparative Study of Histological and Histochemical Image Processing in Muscle Fibre Sections of Broiler Chicken

“…The methods of typing muscle bers are highly advanced and developed, starting by manual typing technics [32], then by interactive tools, which measure the optical density of the bres using spectro-photo-microscopy [27,33], and nally, skeletonization approaches of the inter-bre network in different serial sections, which classify bres according to the gray levels of muscle cells [26,34]. Recently, there was an emergence of automated tools that classify the bres according to their densitometry by superimposing various serial sections coloured differently on a reference image [6,14].…”

A Survey on Medical Image Segmentation Methods with Different Modalitites

“…By separating the entire image into numerous in part covered tiles and disseminating them onto distinctive specialists, simultaneous cell division can be accomplished utilizing an ace laborer way in the Spark distributed computing stage [67]. Our future work is to execute the proposed strategy with distributed computing methods so it can be versatile to substantial scale images.…”