“Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks”の学習報告

The PASCAL Visual Object Classes Challenge

International Journal of Computer and Applications

Artificial intelligence (AI) algorithms, particularly deep learning, have demonstrated remarkable progress in image-recognition tasks. Methods ranging from convolutional neural networks to variational autoencoders have found myriad applications in the medical image analysis field, propelling it forward at a rapid pace. Historically, in radiology practice, trained physicians visually assessed medical images for the detection, characterization and monitoring of diseases. AI methods excel at automatically recognizing complex patterns in imaging data and providing quantitative, rather than qualitative, assessments of radiographic characteristics. In this Opinion article, we establish a general understanding of AI methods, particularly those pertaining to image-based tasks. We explore how these methods could impact multiple facets of radiology, with a general focus on applications in oncology, and demonstrate ways in which these methods are advancing the field. Finally, we discuss the challenges facing clinical implementation and provide our perspective on how the domain could be advanced.

/pdf/artificial-intelligence-in-radiology-2c2bdk1z41.pdf

Artificial intelligence in radiology

Compare your culture to one of the cultures discussed in this unit. On a sheet of paper, list the cultures you are comparing and make one column titled “similarities,” and a second column titled “differences.” Now, list as many similarities and differences between the two as you can think of. Are there more similarities or differences between the two cultures you selected? Have you ever met anyone from this culture? How can you use this information to build greater respect between cultures?

Similarities and Differences

The presence of surgical smoke in laparoscopic surgery reduces the visibility of the operative field. In order to ensure better visualization, the present paper proposes an unsupervised deep learning approach for the task of desmoking of the laparoscopic images. This network builds upon generative adversarial networks (GANs) and converts laparoscopic images from smoke domain to smoke-free domain. The network comprises a new generator architecture that has an encoder-decoder structure composed of multi-scale feature extraction (MSFE) blocks at each encoder block. The MSFE blocks of the generator capture features at multiple scales to obtain a robust deep representation map and help to reduce the smoke component in the image. Further, a structure-consistency loss has been introduced to preserve the structure in the desmoked images. The proposed network is called Multi-scale DesmokeNet, which has been evaluated on the laparoscopic images obtain from Cholec80dataset. The quantitative and qualitative results shows the efficacy of the proposed Multi-scale DesmokeNet in comparison with other state-of-the-art desmoking methods.

Unsupervised Desmoking of Laparoscopy Images Using Multi-scale DesmokeNet

Forearm amputation abruptly affects the patient’s life by limiting their body functionality. Available myoelectric prosthesis somewhat can regain the lost capability of amputees. Nevertheless, there are some limitations associated with these prosthetic devices: (1) their price is excessive (2) their function mainly depends on electromyography (EMG) signals, which are quite susceptible to sweat, motion artifact, electrode shift, and other electrical interference (3) these have sophisticated hardware as well as the control system. This paper introduces an affordable transradial prosthesis controlled by the force myography (FMG) signal. In this work, a unique FMG sensor was designed for the reliable detection of muscular contractions from the remaining forearm of amputees. The sensor was fabricated using a unique mechanical assembly and specific signal conditioning circuitry. A 3D printed prosthetic hand was prepared with an individual position control strategy that receives input from the designed sensor. The designed sensor was validated by determining characteristics such as sensitivity, repeatability, hysteresis, and frequency response. Moreover, its ability to detect muscular contractions was compared with that of an EMG sensor. The designed sensor showed a good correlation (r>0.88) and higher SNR values (>42 dB) as compared to the EMG sensor. Furthermore, the developed hand prototype with the implemented control scheme was successfully verified on five amputees for performing various grasping activities. The amputees were able to control the grasping force of hand fingers with the intent of muscle contraction. The hand offered fast and intuitive operation with input from the FMG sensor.

An affordable transradial prosthesis based on force myography sensor

This paper describes the effect of various glucose concentrations in normal and diabetic human blood serums mixed with intralipid phantoms. Amplitude modulated ultrasound and infrared techniques were utilized here. Modulated ultrasound causes signature specific molecular vibrations through the area of its propagation. The infrared light of specific wavelength is used here to detect the target glucose molecules. The infrared photo sensor detects the resultant signal and undergoes series of signal processing steps to yield the output value. The trend obtained from the result indicates that human diabetic blood serum occupies highest peak values in Fast Fourier Transform domain as compared to the normal human blood serum peak values. The proposed method provides a new dimension for noninvasive blood glucose measurement techniques. Noninvasive method for blood glucose detection is the need of hour. Tight control regimen with noninvasive glucometer will change the way of living of the diabetes affected population.

Measurement of glucose by using modulating ultrasound with optical technique in normal and diabetic human blood serum

Leukemia detection using computer vision algorithms is a significant step in computer-assisted diagnosis for a pathologist. In order to extract the blood cells, many color space models are used for image enhancement and as the preprocessing steps. The present work compares the effect of the green, saturation, Cb and M component of RGB, HSV, YCbCr and CMY color spaces for segmentation of nucleus of blast cells in a leukemia patient's blood smear. The segmentation result of each color space for every ten images is divided into three categories i.e. only WBC segmentation, WBC with peripheral cells and all blood cell segmentation. The study demonstrates that the performance of segmentation is negatively correlated with contrast and illuminance of the input image. HSV and CMY models obtained 85% segmentation accuracy. The present study would help researchers to narrow down their selection when choosing a color space model for segmenting the nucleus of leukemia blast cells.

Automated Nucleus Segmentation of Leukemia Blast Cells : Color Spaces Study

The segmentation of cancerous region is most crucial task in biomedical image processing. The detection of cancerous tissues in early stage is essential for planning the treatment of cancer. In this manuscript, we proposes a modified deep learning architecture for efficient segmentation of biomedical images. Our proposed network effectively Figure out the boundaries of the cancerous region. The network once trained produced exceedingly good results on other datasets without retraining. The evaluation metrics mIoU and BF score shows an improvement of 2.5% and 3% for brain tumour segmentation when trained network is tested on other dataset. Improvement of 2% in both the metrics was achieved for skin lesion segmentation when trained network was tested on other dataset for 100 epochs of training.

Neeraj Sharma

Papers

Unsupervised Desmoking of Laparoscopy Images Using Multi-scale DesmokeNet

An affordable transradial prosthesis based on force myography sensor

Measurement of glucose by using modulating ultrasound with optical technique in normal and diabetic human blood serum

Automated Nucleus Segmentation of Leukemia Blast Cells : Color Spaces Study

Augmented Deep Learning Architecture to Effectively Segment the Cancerous Regions in Biomedical Images