YOLO Based Breast Masses Detection and Classification in Full-Field Digital Mammograms.

BACKGROUND AND OBJECTIVE: With the recent development in deep learning since 2012, the use of Convolutional Neural Networks (CNNs) in bioinformatics, especially medical imaging, achieved tremendous success. Besides that, breast masses detection and classifications in mammograms and their pathology classification are considered a critical challenge. Till now, the evaluation process of the screening mammograms is held by human readers which is considered very monotonous, tiring, lengthy, costly, and significantly prone to errors. METHODS: We propose an end to end computer-aided diagnosis system based on You Only Look Once (YOLO). The proposed system first preprocesses the mammograms from their DICOM format to images without losing data. Then, it detects masses in full-field digital mammograms and distinguishes between the malignant and benign lesions without any human intervention. YOLO has three different architectures, and, in this paper, the three versions are used for mass detection and classification in the mammograms to compare their performance. The use of anchors in YOLO-V3 on the original form of data and its augmented version is proved to improve the detection accuracy especially when the k-means clustering is applied to generate anchors corresponding to the used dataset. Finally, ResNet and Inception are used as feature extractors to compare their classification performance against YOLO. RESULTS: Mammograms with different resolutions are used and based on YOLO-V3, the best results are obtained through detecting 89.4% of the masses in the INbreast mammograms with an average precision of 94.2% and 84.6% for classifying the masses as benign and malignant respectively. YOLO's classification network is replaced with ResNet and InceptionV3 to get overall accuracy of 91.0% and 95.5%, respectively. CONCLUSION: The proposed system showed using the experimental results the YOLO impact on the breast masses detection and classification. Especially using the anchor boxes concept in YOLO-V3 that are generated by applying k-means clustering on the dataset, we can detect most of the challenging cases of masses and classify them correctly. Also, by augmenting the dataset using different approaches and comparing with other recent YOLO based studies, it is found that augmenting the training set only is the fairest and accurate to be applied in the realistic scenarios.

Hybrid, randomized and high capacity conservative mutations DNA-based steganography for large sized data.

In this paper, a well secured, high capacity, preserved algorithm is proposed through integrating the cryptography and steganography concepts with the molecular biology concepts. We achieved this by first encrypting the confidential data using the DNA Playfair cipher to avoid extra information sent to the receiver and it consequently acts as a trap for an attacker. Second, it achieves a randomized steganography process by exploiting the DNA conservative mutations. The DNA conservative mutations are utilized in a way that allows a DNA base to be substituted by another base to allow carrying two bits. Consequently, a high capacity feature is obtained with no payload for the used sequence. There are three main achieved contributions in this work. First, is hiding high capacity of data within DNA by exploiting each codon to hide two bits whilst preserving the sequence properties of protein after the steganography process, which is a trade off in the field. Secondly, using the conservative mutation with all its valid biological permutations, leads to the lowest cracking probability achieved and published till now, as proven in the security analysis section. Finally, a comparison is conducted between the proposed algorithm and five recent substitution based algorithms using large sized data up to three megabytes, to prove the algorithm's scalability.

Color image segmentation based on different color space models using automatic GrabCut.

This paper presents a comparative study using different color spaces to evaluate the performance of color image segmentation using the automatic GrabCut technique. GrabCut is considered as one of the semiautomatic image segmentation techniques, since it requires user interaction for the initialization of the segmentation process. The automation of the GrabCut technique is proposed as a modification of the original semiautomatic one in order to eliminate the user interaction. The automatic GrabCut utilizes the unsupervised Orchard and Bouman clustering technique for the initialization phase. Comparisons with the original GrabCut show the efficiency of the proposed automatic technique in terms of segmentation, quality, and accuracy. As no explicit color space is recommended for every segmentation problem, automatic GrabCut is applied with RGB, HSV, CMY, XYZ, and YUV color spaces. The comparative study and experimental results using different color images show that RGB color space is the best color space representation for the set of the images used.