Detection of breast cancer has constantly been a key problem for the clinicians and physicians for disease analysis and proposing the right therapy. The physical diagnosis of breast cancer from histopathology images is highly subjective and may differ from person to person subject to their expertise level. Also, the manual identification depends on other factors like precise environment and lighting arrangement. The computer assisted diagnosis systems helps in lessening the aforementioned problems and offers improved outcomes in detecting the breast cancer from histopathology images.
The combined recent advancements in the field of medicine and engineering give a way for curing this cancer. Early detection is the only answer for curing the breast cancer. The stage of malignancy determines the choice of treatment of breast cancer. A biopsy is a clinical procedure in which a tissue from the breast is removed and examined under a microscope for detecting suspicious cells. Thus, analysis of this biopsy image is a vibrant method for detecting breast cancer. Histopathology analysis is the study of signs and warnings of the breast cancer from the biopsy images. A procedure called staining is performed on these images to clearly visualize the structures and nuclei in the tissue. The tissue undergoes staining process using various dyes. The pathologists were using Hematoxylin & Eosin (H&E) staining from past few years. Hematoxylin is responsible for blue color in cell (nuclei) whereas Eosin is responsible for pink color in cytoplasm and other connective structures. Abnormal and normal tissues can be differentiating from histopathology images with high resolution.
Previous literature exhibits a massive collection of breast cancer detection devices based on FPGA implementation. FPGAs are better for prototyping and low quantity production. When the quantity of FPGAs to.be manufactured increases, cost per product also increases. This is not the case with ASIC implementation. But FPGA is a chip not a board so more development is needed there. But Raspberry Pi comes with an OS, a suite of open source image processing applications even an interface to a camera build in.
There are many researches were conducted on automated detection and classification of breast cancer from histopathology images. All these studies extracted some important features from the regions of interest (ROI) and evaluate the possibility of cancer among the patients. This was achieved through classification of images into benign and malignant cancers. The present invention comprises of a smart breast cancer detection device based on Raspberry Pi and a technique using several machine learning algorithms for automated classification of breast cancer from histopathology images. This invention is particularly applicable for medical imaging applications.
SUMMARY OF THE INVENTION
[003] The invention and further developments and modifications of the invention have been defined in the patent claims. The present realisation is a complete integrated system for detection of breast cancer. The device is skilled

in detecting breast cancer by scanning histopathology images by using a camera. The device is fully packed in Raspberry Pi board. A camera is attached to the board for scanning the images. The board is working as a mini computer which is programmed with various machine learning techniques. The final decision analysis is displayed by the device with the help of an LCD display. This BC scanner can be proven to be a valid choice for clinicians to use it as an alternate option. This device can provide commercially to create a low-cost effective patient breast cancer monitoring system.
A principal feature of this invention includes the Raspberry Pi board and it contains a processor and graphics chip, program memory (RAM) and various interfaces and connectors for external devices. This invention comprises of a CPU named BCM2835 which is cheap, powerful, and it does not consume a lot of power. Raspberry Pi operates in the same way as a standard PC, requiring a keyboard for command entry, a display unit and a power supply. SD Flash memory card normally used in digital cameras is configured in such a way to 'look like' a hard drive to Raspberry Pi's processor. The unit is powered via the micro USB connector. Internet connectivity may be via an Ethernet/LAN cable or via an USB dongle (Wi-Fi connectivity).
Another optional feature during the histopathology image analysis is the colour normalisation module. The images taken by a CCD camera should be normalized since there will be inconsistencies due to light and stain variation. The normalization approach consists of two modules: illuminant normalization and stain normalization. These two modules can be performed either independently or one after the other. Prior to analysis, all the input images were pre-processed for better analysis. Main issues facing by histopathological images are noises, changes in resolution and weak contrast. This leads to difficulty in cancer detection. So, some pre-processing steps are required to mitigate these problems and make feature extraction easier.
Yet another feature employs techniques like the pre-processing step which consists of color normalization and image enhancement to improve the image clarity. Then a segmentation procedure using A-means clustering has done followed by hybrid feature extraction phase. The features that are extracted involve geometrical features, color features and texture features. Next, the breast cancer images are classified using Support Vector Machine, k-Nearest Neighbor, Random Forest and Artificial Neural Network. Finally, the Raspberry Pi was programmed with ANN for classifying benign and malignant cancers.
Consistent with some embodiments, there is provided a GSM module and a cloud computing to connect with doctor's smart phone and hospital server respectively. When the doctor's phone receives the decision outcome, he personally contacts the patient. At the same time, there is included a provision of getting result to patient's mobile also.
LIST OF PREFERRED AND OPTIONAL FEATURES
1. The proposed device comprises of a Pi camera for scanning histopathology breast images and an LCD

display for displaying the predicted decision outcome: benign or malignant lesions. 2.The device incorporates a GSM module with cloud computing that can be interfaced with smart phone of both doctor and patient and also to the hospital server for monitoring on a regular basis.
BRIEF DESCRIPTION OF THE DRAWING
The following figures are used to explain the concepts and design of the breast cancer detection system of the
invention: Figure 1 illustrates the workflow process to obtain a histopathology image used by the proposed device via biopsy. Figure 2 shows the concept design of the breast cancer detecting device. Figure 3 depicts the block diagram of the system architecture of the present invention. Figure 4 shows the offline analysis of breast cancer detection. Figure 5 is a flowchart showing the machine learning based classification system for detecting breast cancer,
consistent with some embodiments. Figure'6 shows the normalisation result of H&E stained histopathology breast image. Figure 7 depicts the segmentation result with A-means algorithm.
Figure 8 shows the validation analysis performance curve for the neural network used in the present device. In the drawings, elements having the same designation have the same or similar functions.
DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS
[004]" Certain embodiments are described in detail with certain specific details in the following description. It will
be apparent, however, to one skilled in the art that the disclosed embodiments may be practiced without some or all of these specific details. There are no limitations for the specific embodiments presented here but they are meant to be illustrative. One skilled in the art may realize other material that, although not specifically described herein, is within the scope and spirit of this disclosure. This invention utilizes a number of machine learning (ML) techniques with different learning methods in a computer-aided detection, analysis and diagnosis (CAD) device. To understand the invention and its novel features, the basic concept will first be briefly described. Next, the ability of the ML algorithms will be explained. Then finally, the Raspberry Pi smart invention will be demonstrated in

detail with its capabilities.
Figure 1 illustrates the basic concept of biopsy process through which a histopathology image obtained. Biopsy is a surgical process where a tissue is removed from any part of the body and examines for the existence, reason and scope of the cancer. This process is usually carried out in a histology lab by a pathologist. Histopathology refers to the process of studying specimen of tissue by a pathologist under the microscope for the detection of cancer. Different types of staining exist. Hematoxylin-Eosin (H&E) staining is a popular staining technique to visualize structures in tissue samples. H&E staining represents histological structures in a logical and distinct way. Tissue nuclei will be coloured as blue or violet and eosinophilic texture will be coloured as red or pink. These coloured specimens are examined by the pathologist under microscope with a CCD camera.
Figure 2 demonstrates the design of overall concept of the present invention. As shown in Fig. 1, the histopathology image was captured by a CCD camera mounted on the microscope. So, from the patient, the histopathology image is obtained. Later, as in Figure.2, the image is scanned by a Pi camera mounted on Raspberry Pi board and analysed , by the device which is already programmed in Python. Consistent with some embodiments, the images were , classified into benign and malignant using ANN. The decision result that is either benign or malignant will be displayed in an LCD display. Along with this, a GSM module is connected to pass information to the doctor's smart phone and Patient's smart phone. A cloud system is also included for accessing decision result from the proposed device by the hospital server thereby connected to doctor. Here, an extra option included to give a feedback from doctor to patient personally. Figure 3 illustrates the block diagram of the present invention. Explanation is same as given by Figure.2.
The breast cancer detection analysis pipeline is shown in Figure.4. As shown in Figure.4, the present invention comprises of a study that consists of three steps: raw data pre-processing, image data analysis and classification. Preceding to classification, the pre-processing stage consists of colour normalization and image enhancement to improve the image clarity. Then a segmentation procedure using A-means clustering has done followed by hybrid feature extraction phase. The features that are extracted involve geometrical features, colour features and texture features. Next, the breast cancer images are classified using SVM, £-NN, RF and ANN. Then the performance

metrics were evaluated and compared to find out the most accurate classifier.
The detailed block diagram of classification comparison module has illustrated in Figure 5. The final step in any cancer analysis pipeline is the classification task. This is one of the most challenging tasks in cancer detection. There are supervised and unsupervised learning approaches for classification and regression problems. Classification module will predict whether the histopathology image is a benign one or malignant one. The various classifiers available are SVM, £-NN, Bayesian classifier, Convolution neural network, ANN and so on. Initial stage is to formulate the feature set, the next stage is to pick a suitable algorithm for evaluation, the third stage is model fitting, the fourth stage is the training stage in which trains the fitted model, and then the last stage is to utilize this model (fitted) for predicting the outcome. The ^-nearest neighborhood (KNN), support vector machine (SVM), artificial neural network (ANN) and random forest classifiers are exercised for predicting the benign and malignant cancer images. Consistent with some embodiments, Raspberry Pi was programmed with ANN to classify the histopathology images into benign and malignant.
Figure 6 illustrates the colour normalisation output and the input image. The colour normalisation approach was included to mitigate the colour difference problem caused by both light and stain concentration variations. The normalization approach consists of two modules: illuminant normalization and stain normalization. These two modules can be performed either independently or one after the other. The Power Spectral Density (PSD) of the imaging light, P(y) has to be known for the algorithm to perform. P(y) affects the intensity of the image, I(n,yi) where I(n,yi) varies with stain spectra, S(y) exponentially. Light variation during biopsy imaging of a specimen causes various color biases in the image. Estimating P(y) of the query image and matching it with Ps(y), PSD of standard image will help to reduce color bias. Ps(y) can be accessed from either reference image or can be predefined. Light normalization has two parts viz: Light matching and Light estimation. While staining a biopsy tissue using chemical dyes, chances are there for getting a particular part more stained. Also, stains may get blended together. Then pathologists may get mislead and analysis will be difficult. Due to this, stain spectra, Si(y) changes which leads to color variation. The stain normalization module mitigates this color variation. The histological information is maintained by D(m). Normalization of Si(y) by preserving data on D(m) is conveyed by

using an NMF. based stain estimation and matching.
Figure 7 shows the segmentation result of the input histopathology image using k means algorithm for extracting
the ROI from which the features are going to extract. A-means algorithm is a subdividing method that partition a set
of data into k groups. The algorithm moves iteratively through the following steps. Evaluate the mean of each and
every group or cluster. .Then estimate the distance of each data point from each group to the cluster centre. Finally,
allocate each data point to the nearest cluster on the basis of distance calculated. Once the allocation is over,
recalculates the cluster centre again and based on that centre, new distance vector has to measure.
Imagine an x x y image that has to be partition into k clusters and c* be the centre for clusters. Then A-means
I algorithm is as follows: Set the number of clusters, k and its centre.
, Calculate the distance between each data point in the image and the cluster centre by using the equation given
below:
d = \p{x,yyck\ ^
Allocate all the data points into clusters based on the nearest distance, d.
Once allocation is done, recalculate the new centre for every cluster using the equation given below:
Repeat the procedure until it converges.
Assign data points again accordingly and reshape the image.
Figure 8 illustrates the validation analysis performance curve for artificial neural network that was used in
Raspberry Pi for classifying images. A neural network classifier that uses back propagation algorithm (BPA) is
used here. It is a connection of neural processors in which one node is connected to other and they exchange
outputs from one node to next. Generally, the activation functions that are used in BPA are binary and bipolar
sigmoid functions. Classification part consists of two stages: training and testing. During training, a set of

predefined images with their output class is fed to the network for learning. Then a set of new images (test images)
have been tested for classification. Now the neural network classifies the test images according to its knowledge
based on training. The activation function is given by,
A(x) = l/(l+e*(-x)) (3)
The output layer weight update is given by,
New weight = previous weight + learning rate * activation value from the previous node * (Expect Output - Actual
Output) * A'(input to activation function1 of that node) (4)
and the hidden layer weight update is given by,
New weight = previous weight + learning rate * activation value from the previous node * A'(input to activation
function of that node) * Sum for each node in the next layer (weight of this node to output node * Error of output
node * A'(input to output node's activation function)) (5)
[005]The proposed system involves an efficient technique to detect such HTTP flooding DDoS attacks in the cloud.
This proposed method is integrated into the cloud ecosystem so that the attacks from different attack paths are
covered to safeguard the targeted cloud application along with complete cloud environment. This technique
involves an advanced zonal based classification protection mechanism with three zonal areas. The first one is the
Initial Monitoring Zonal Area (IMZA), where the incoming HTTP requests are monitored to classify the client
based on their behavior.
[006] The requests are identified as legitimate, they are classified into Allowed Green Zonal Area (AGZA), where
the requests are serviced continuously. If the requests are identified as flooding attack, classified int Blocked Red
Zonal Area (BRZA) so that none of the requests is processed further. In the case, if the initial requests seem to be
legitimate and classified into AGZA, later starts the flooding attack the proposed system identifies the same and
moves it into BRZA to make sure application and the cloud environment are safeguarded."