Determining the Degree of Malignancy on Digital Mammograms by Artificial Intelligence Deep Learning

Abstract

In this paper, we propose a method for determining degree of malignancy on digital mammograms using artificial intelligence deep learning. Digital mammography is a technique that uses a low-energy X-ray of approximately 30 KVp to examine the breast. The goal of digital mammography is to detect breast cancer in an early stage by identifying characteristic lesions such as microcalcifications, masses, and architectural distortions. Frequently, microcalcifications appear in clusters that increase ease of detection. In general, larger, round, and oval-shaped calcifications with uniform size have a higher probability of being benign; smaller, irregular, polymorphic, and branching calcifications with heterogeneous size and morphology have a higher probability of being malignant. The experimental images for this study were selected by searching for "mammogram" in the NIH database. The images were converted into JPEG format of 256 X 256 pixels and saved. The stored images were segmented, and edge detection was performed. Most of the lesion area was low frequency, but the edge area was high frequency. DCT was performed to extract the features of the two parts. Similarity was determined based on DCT values entered into the neural network. These were the findings of the study: 1) There were 6 types of images representing malignant tumors. 2) There were 2 types of images showing benign tumors. 3) There were two types of images demonstrating tumors that could worsen into malignancy. Medical images like those used in this study are interpreted by a radiologist in consideration of pathological factors. Since discrimination of medical images by AI is limited to image information, interpretation by a radiologist is necessary. To improve the discrimination ability of medical images by AI, extracting accurate features of these images is necessary, as is inputting clinical information and accurately setting targets. Study of learning algorithms for neural networks should be continued. We believe that this study concerning recognition of cancer on digital breast images by AI deep learning will be useful to the radiomics (radiology and genomics) research field.

Spatial Frequency Analysis by DWT of CXR in the COVIDGR Dataset

“…The COVIDGR-19 Feature Data were entered into an Excel file to calculate the sum, and the average value was obtained and confirmed as the "Feature value of images by disease. [11][12] " A frequency (Approximation, Horizontal, Vertical, Diagonal) analysis was performed [10].…”

Review: Application Trends of Contrast Media in PET/MRI

“…Microbubbles [31 ] Microbubbles 18 F Non-invasive angiography 64 Cu-labledmagneticnanoparticle [32] SPIO 64 Cu Atheromatous palque imaging 124 I-SA-MnMEIO [33] SPIO 124 I Lymph node imaging RGD-PASP-IO combined with 64 Cu [34] PASP-IO 64…”

“…Tumor neoangiogenesi s imaging 18 F-FDG labeled MNPs [34] Iron oxcid 18F molecular imaging [35] . To avoid these disadvantages, iron oxide nanoparticles that are monodisperse and of uniform crystallinity were prepared using a thermal decomposition method.…”