Liver Cancer Detection and Grading using Efficient Computer Vision Techniques
Mr. Shaikh Imran Turab*, Dept. of Electronic & Telecommunication Eng. Govt.
Dr. V. K. Kadam, Professor, Dept. of Electronics & Telecommunication, P.E.S. College of Eng. Aurangabad
Abstract: Early prediction of any kind of cancer always advantageous for on-time medical treatment to save the patient lives. The Computer Aided Diagnosis (CAD) tools using signal processing and image processing methods gained significant attentions for immediate and accurate diagnosis using patient’s raw medical data like Magnetic Resonance Imaging (MRI), Chromatography (CT), etc. The liver cancer early detection and analysis of its grading is important research problem. This paper proposed the CAD system for early detection liver cancer accurately followed by its grading analysis into different stages like stage 1 (T1), stage 2 (T2), and stage 3 (T3). The proposed framework consists of stages like pre-processing, Region of Interest (ROI) extraction, features extraction, and classification. The raw CT scans of liver pre- processed to remove the noises using the filtering and contrast adjustment functions. The adaptive segmentation method designed to using binarization and morphological operations to extract the accurate ROI with minimum computational burden. For features extraction, the text and shape features extracted using Gray Level Co-occurrence Matrix (GLCM) and geometric moment methods respectively. The conventional classifiers such as Artificial Neural Network (ANN) and Support Vector Machine (SVM) applied for prediction. The experimental results shows accuracy of proposed model improved existing methods.
Keywords: Artificial neural network, computer aided diagnosis, adaptive segmentation, classification, features extraction, liver cancer detection.
Cancer is the significant danger for person wellbeing and its number of patients expanding word wide because of the an unnatural weather change, regardless of whether there are new treatments and therapies proposed by research specialists, however level of cancer characterizes the capacity of its fix. There are various kinds of cancers from which individual is enduring male and female.
Computers have been successfully applied to various fields of medical sciences such as biochemical analysis, drug development and recognition of diseases from medical images .
Successful identification of lung cancer, brain tumor is possible with the existing CAD. However, little research has been focused on liver because of the difficulties in segmenting liver from other adjacent abdominal organs such as kidney, stomach and gall bladder using abdominal images due to gray level similarities of adjacent organs. The most well-known clinical imaging reads for early discovery and analysis of liver illnesses incorporate Ultra Sonography (US), Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) .
Liver illnesses are paid attention to, in light of the fact that liver is indispensable critical to the existence of a patient. Liver is perhaps the biggest organ in the human body situated in the upper right bit of the midsection. The liver has numerous significant capacities, such as clearing poisons from the blood, utilizing drugs, blood proteins and produce bile which helps absorption . Liver can be forever harmed because of various reasons which incorporate infection contaminations, response because of medications or liquor, tumors, genetic conditions and issue with the body's invulnerable framework. Liver sicknesses establish a significant clinical issue of overall extents.
Approximately 50% of the people  are affected by liver diseases.
Liver diseases are mainly classified into diffused liver diseases and focal liver diseases based on the dispersion in the pathology. Diffused liver diseases are distributed throughout the whole liver volume whereas Focal liver diseases are concentrated in small spots in one or both of the liver lobes while the rest of the liver tissues remain normal  . A doctor may diagnose a disease on the basis of symptoms, laboratory test results, patient's medical history, physical examinations and scan reports. For example, during physical examination, the doctor may notice that the liver is harder or larger than usual and order blood tests that can show whether the disease is present. The doctor can ask for a scan if it is needed. Central liver sores are as often as possible identified in patients going through stomach examinations. The liver tumors comprise a significant analytic test for radiological imaging, particularly when cancer patients are included. Most benevolent tumors are found by chance on an imaging investigation of the liver, for example, ultrasound or CT examine. Sometimes, a biopsy might be needed to make the finding of hepato cell adenoma . Harmful tumors might be identified by screening high danger patients or by chance on an imaging investigation of the midsection performed for another explanation or might be distinguished on account of side effects, for example, stomach torment. In patients, who experience the ill effects of further developed hepatocellular carcinoma, weight reduction, intermittent serious agony and other summed up indications may happen. The analysis of hepato cell cancinoma is ordinarily made by liver imaging tests, for example, stomach ultraround and CT filter in mix with the estimation of blood levels of aiphafeto protein. The existing tests such as
biopsy are conducted for the final diagnosis of liver cancer. Such tests are difficult and expensive as the attention of experienced doctor required to analysis the CT scan images. Therefore in recent studies, Computer aided design (Computer Aided Diagnosis) can help radiologist and doctors in identifying sores and in separating kindhearted and dangerous injuries on clinical pictures. The outcomes got from CAD can be utilized as a "second assessment" by radiologists in their understandings which improve indicative precision . Various CAD plans have been produced for discovery and arrangement of sores in clinical pictures. Execution examines show that the computer yield assisted radiologists with improving their indicative precision. As, CAD can be applied to all imaging modalities, all body parts and a wide range of assessments, almost certainly, CAD will significantly affect clinical imaging and symptomatic radiology in the 21St century  . However, the accuracy of liver cancer detection and analysis is mainly depends of accurate estimation of Region of Interest (ROI) using the effective image segmentation methods. This paper proposed the efficient CAD tool for early prediction of liver tumor detection and automatic grading of its severity. Section II presents the review of various recent methods related to the same domain. Section III presents the design of proposed model. Section IV presents the simulation results and analysis. Finally conclusion discussed in section V.
II. Related Works
Several image processing based techniques proposed using various approaches and classification techniques. The core focuses of such methods on tumor segmentation, features extraction, and classification phases. Some of the recently introduced related works described in this section.
In , creator proposed programmed division of liver injuries utilizing half breed division methods. Their framework depended on two distinctive datasets and test results indicated that the proposed framework was quick, strong and viable in recognizing the presence of injuries in the liver, and figure the region of liver influenced as tumors sore, and gave order precision of 93%, which could section liver and concentrate sores from stomach CT in under 0.15 s/cut.
In , author designed an accurate method for liver tumor segmentation, in which the noise removed liver CT image by preprocessing was employed with SVM classifier for tumor segmentation. The SVM was already trained using the manually given image sets before test image classification. One after the other, feature extractions and morphological operations were performed on the binary image for further refinement ofthe rough segmentation result of SVM classifier. Their experiment results provided an accuracy of 97.83% for dataset 1 and 95.23% for
In , author predicted the global shape properties and pose of liver tumor regions with a statistical shape model to study the shape features of the liver from the liver CT image. In their method, a template was used to recover local deformations by comparing the template with the original liver image. The experiment results were validated for 10 cases and obtained moderate results.
In , author proposed quick, conventional calculation for estimating the volume of strong, minimal tumors in CT that considers halfway volume impacts at the boundary of a given division result. It is an expansion of the division based incomplete volume examination technique.
In , author announced arbitrary walker based structure that can section contrast-upgraded livers CT pictures with extraordinary precision and speed. In view of the area of the correct lung projection, the liver arch consequently distinguished in this manner wiping out the requirement for manual introduction. The computational prerequisites are additionally limited using rib- confined zone division; the liver is then extricated by using irregular walker technique.
In , author acquainted a robotized CAD framework with characterize liver sores into Benign or Malignant. The framework comprises of three phases; initially, programmed liver division and sore recognition. The highlights are extricated from various ROIs, and afterward ordering liver sores into kind and harmful. The technique separates a divided injury into three zones, for example inside, outside and line zones.
In , proposed programmed liver tumor division from stomach CT check pictures. A measurable boundary based methodology is utilized to recognize liver tumor tissue from other stomach organs. The current division strategies, for example, area developing and force based thresholding techniques examined.
In , author proposed and assessed a novel programmed approach for Oral Squamous Cell Carcinoma diagnosis utilizing profound learning advancements on CLE pictures. The technique is analyzed against textural include based AI moves toward that address the present status of the craftsmanship.
In , author introduced comparative and various tissues in CT pictures of liver are dictated by utilizing two unique techniques; watershed and histogram thresholding. The pictures have been pre-prepared before division.
In , author proposed multi-modular picture investigation for assurance of the liver and perfusion domains. They utilized the self-loader liver division approach for ROI extraction.
In , author proposed the a programmed approach that coordinates the versatile thresholding and spatial fluffy bunching approach for location of cancer district in CT filter pictures of liver.
They assessed the proposed calculations on 123 CT examine pictures.
In , author proposed a staggered outfit model to identify a liver and its order utilizing highlights extricated from CT pictures. The k-overlap cross-approval applied to legitimize the vigor of the classifiers. Contrasted with the individual classifier, the staggered outfit model accomplished high exactness in both the recognition and characterization of various tumors [26- 31].
This section presents the design of proposed model of detecting the liver cancer and grading analysis of tumor severity using the extracted features. Optimized CAD tool for early liver cancer detection using image processing terminologies is main goal of this paper. The practical approach used for evaluating the efficiency of proposed liver cancer detection method. The methods such as bilateral filter for pre-processing, ROI segmentation using the binarization and morphological operations, hybrid features for disease grading analysis and detection designed as showing in figure 1.
Figure 1. Proposed CAD model for liver cancer detection and tumor analysis
Figure 1 demonstrating the functionality of detecting the liver cancer. If liver cancer detected in input CT image, tumor analysis performed to discover the stage of cancer. The functionality of each phase of proposed model elaborated in below sections.
A. Pre-processing: As the quality of input liver CT scans may contain noises, it can lead to suppressions of tumor areas and hence possibility of wrong predictions. The recent automated deep learning based techniques designed without image pre-processing. These methods may not be reliable for a longer time. The input CT image C pre-processed in the proposed model by applying techniques that work adaptively such as intensity values adjustment and median filtering.
The first operation focused on adjusting the image intensity values of low contrast CT images as:
𝐶1 = 𝑖𝑚𝑎𝑗𝑢𝑠𝑡 (𝐶) (1)
Where, 𝐶1 is outcome of contrast enhancement step using function 𝑖𝑚𝑎𝑑𝑗𝑢𝑠𝑡.
After adjusting the image contrast, we applied the 2D filtering method to suppress the artefacts and noises effectively. The median filtering used to remove the noises in contrast enhanced image. It may possible that adjusting the image intensity values leads to noise and also X-ray scan also introduce the noises in image. The 2D median filtering works by moving via the image pixel by pixel, replacing every value with the neighbouring pixels median value. The neibhours pattern is decided by the size of window. The window size of 3-by-3 neighborhood is used in this work.
The 2D median filter is applied on 𝐶1 as:
𝐶2 𝑖, 𝑗 = 2𝐷𝑚𝑒𝑑𝑖𝑎𝑛 𝐶1 𝑖, 𝑗 𝑖, 𝑗 ∈ 𝑤 (2)
Where, 𝐶2 is outcome of median filtering and 𝑤 is the size of window.
B. ROI Extraction: After the task of pre-processing, image segmentation method used to extract the image ROI for further analysis. The extraction of tumor related information from the pre- processed image 𝐶2 accurately is important research problem. The conventional techniques suffered from challenges like inaccuracy, over-segmentation, etc. In this paper, we designed the robust but accurate ROI extraction technique using binarization followed by morphological operations. The binary image segmentation is defined as the approach of classifying the intensity values of skull images into foreground regions and background regions using the threshold value.
The threshold value computed dynamically for each input pre-processed skull image using Otsu’s  technique. To improve the accuracy of binary segmentation, we applied morphological operations. The steps of proposed segmentation are:
- Compute dynamic threshold value of input pre-processed image 𝐶2 - Apply binary segmentation using computed threshold value of 𝐶2
- Apply morphological structuring element operation using disk size 3 on segmented image - The structuring element used in morphological closing operation to produce the accurate
ROI image 𝐶3 - Return 𝐶3
C. Features Extraction and Normalization: The ROI image represented by the unique values called as the features in image processing terminology. The features extraction is another vital phase of any CAD tool. The rich and unique set of features leads to accurate classification and disease analysis. In this paper, we designed two types of features such as texture features using GLCM and moment invariant features. Both features deal with geometry, shape, and texture properties of ROI images. The well-known GLCM technique used to extract 20 features that consist of 16 GLCM features and 4 statistical features. The 4 GLCM properties such as contrast, correlation, energy, and homogeneity computed to get 16 features. We first compute the GLCM of ROI image using four offset [0 1;-1 1;-1 0;-1 -1:] as:
𝐺𝑚 = 𝑔𝑙𝑐𝑚(𝐶3, 0 1; −1 1; −1 0; −1 − 1 ) (3)
Using 𝐺𝑚, four texture features computed of size 1 × 4 of each. This builds the 1 × 16 feature vector for each input ROI image. Let, 𝐺𝑚 is the GLCM matrix and 𝐿 is maximum possible quantized value. Table 1 shows the texture features with equations.
Table 1. GLCM features
Texture feature Equation Equation Number
|𝑖 − 𝑗⃓2
𝑗 −0 𝐿−1
1 + ⃓𝑖 − 𝑗⃓2
𝑗 −0 𝐿−1
𝜎𝑥𝜎𝑦 𝑖 − 𝑗 𝐺𝑚𝑖𝑗 − 𝜇𝑥𝜇𝑦
𝑗 −0 𝐿−1
𝜇𝑥 = 𝐺𝑚𝑖𝑗
𝜇𝑦 = 𝐺𝑚𝑖𝑗
𝜎𝑥2 = (𝑖 − 𝜇𝑥
𝜎𝑥2 = (𝑗 − 𝜇𝑥
𝑗 −0 𝐿−1
After that, we computed 2D statistical features such as mean, standard deviation, entropy, and variance of ROI image. It form total 1 × 20 texture vector of input ROI image. To represents the shape of ROI, we extracted the 8 geometric moment features from ROI image. The geometric moment of order (𝑝, 𝑞) for a ROI image is computed as:
𝜇𝑝𝑞 = 𝑚 𝑖=1 𝑛𝑗 =1(𝑖 − 𝑖 )𝑞𝐶3(𝑖, 𝑗) (11)
Where, 𝑖 =𝑚10 𝑚00 and 𝑗 =𝑚01 𝑚00 are the coordinates of the object centroid. In this way, we have computed 8 moments as: 𝑚00 = 𝑚 𝑖=1 𝑛𝑗 =1𝐶3(𝑖, 𝑗) (12)
𝑚10 = 𝑚 𝑖=1 𝑛𝑗 =1𝑖 𝐶3(𝑖, 𝑗) (13)
𝑚01 = 𝑚 𝑖=1 𝑛𝑗 =1𝑗 𝐶3(𝑖, 𝑗) (14)
𝜇11 = 𝑚 𝑖=1 𝑛𝑗 =1 𝑖 − 𝑖 (𝑗 − 𝑗 ) 𝐶3(𝑖, 𝑗) (15)
𝜇12 = 𝑚 𝑖=1 𝑛𝑗 =1 𝑖 − 𝑖 (𝑗 − 𝑗 )2𝐶3(𝑖, 𝑗) (16)
𝜇21 = 𝑚 𝑖=1 𝑛𝑗 =1( 𝑖 − 𝑖 )2(𝑗 − 𝑗 ) 𝐶3(𝑖, 𝑗) (17)
𝜇30 = 𝑚 𝑖=1 𝑛𝑗 =1( 𝑖 − 𝑖 )2((𝑗 − 𝑗 ))2 𝐶3(𝑖, 𝑗) (18)
𝜇30 = 𝑚 𝑖=1 𝑛𝑗 =1 ( 𝑖 − 𝑖 )3𝐶3(𝑖, 𝑗) (19)
The texture features and moment features fused to form total 28 features in vector V of each input CT image. As the there is significant variations in these 28 features, we opt for features normalization for performance improvement. The feature normalization convert all features in range of 0 to 1 using min-max normalization technique. The normalized features vector 𝑉norm produced by:
𝑉norm = (𝑉−min 𝑉 )
(𝑚𝑎𝑥 𝑉 −min 𝑉 ) (20)
D. Detection and Classification: The training performed on complete dataset that consists of two categories normal and diseased CT images. For detection, we used two classifiers ANN and SVM.
The performance of these classification methods are investigated by dividing the training data into the ratio of 70 % training and 30 % testing. If classification output is detected as cancer disease, then we applied the severity analysis to estimate the stage of cancer. This can be done by using the 𝑉norm feature vector.
IV. Results Analysis
This section presents the simulation results and comparative analysis of proposed model using MATLAB tool. The dataset consist of 100 CT scan liver images collected from different sources from Kaggle  and Github  repositories. 50 images are normal and 50 are liver cancer of different subjects. The performance of proposed model investigated using ANN and SVM classifiers against the texture features (20 features), moment features (8 features), and normalized- fused features (28 features). The performance metrics such as accuracy, precision, recall, F1- score, and specificity parameters investigated.
Figure 2. ROC representation of proposed model
Figure 3. Confusion matrix representation of proposed model
A figure 2 and 3 demonstrates the Receiver Operating Characteristics (ROC) curve and confusion matrix respectively using ANN classifier for proposed model. It shows the 87 % overall accuracy considering both classes achieved using techniques designed for pre-processing, segmentation, and features extraction.
Figure 4. Accuracy analysis using different classifiers
Figure 5. Precision analysis using different classifiers
Figure 6. Recall analysis using different classifiers
Figure 7. Specificity analysis using different classifiers
Figures 4-8 demonstrates the analysis of accuracy, precision, recall, specificity, and F1-score parameters. As discussed earlier, the individual moment and texture features investigated against the proposed fused and normalized features vector using SVM and ANN classifiers. The accuracy results show that proposed features extraction technique achieved higher accuracy compared to moment and texture features. The similar trend observed for precision, recall, specificity, and F1- score parameters. Among the SVM and ANN classifiers, the ANN classifier achieved higher accuracy of detection compared to SVM. The F1-score and accuracy of proposed model using ANN is approximately 92 % and using SVM is approximately 88 %.
Figure 8. F1-score analysis using different classifiers
Furthermore, we investigated the performance of proposed model with recent techniques introduced in , , and . All three methods similar to proposed model, thus included for comparative study. Table 2 shows the comparative analysis in terms of accuracy and average detection time. The proposed model shows the efficiency and robustness compared to all existing recent techniques. This is mainly because of using simple and accurate technique of ROI extraction and features extraction.
Table 2. State-of-art methods analysis
Methods Accuracy (%) Avg. detection time (seconds)
 86.71 1.93
 88.99 3.41
 91.05 1.43
Proposed 92.67 1.13
V. Conclusion and Future Work
This paper proposed the CAD model for liver cancer detection and automatic cancer severity analysis using the similar set of normalized hybrid features. In this regards, we applied the median filtering for pre-processing, then dynamic threshold-based binary segmentation followed by morphological functions used for ROI extraction. From ROI image, texture and moment features extracted, fused, and normalized. For classification purpose, ANN and SVM classifier applied.
The simulation results prove that proposed model improved the accuracy of detection with minimum computational overhead. For future work, we suggest to introduce the deep learning technique for automatic detection and accuracy improvement.
1. Nader H. Abdel-massieh, Mohiy M. Hadhoud, Khalid M. Amin, "A Novel Fully Automatic Technique for Liver Tumor Segmentation from CT Scans with knowledge- based constraints", 2010 10th International Conference on Intelligent Systems Design and Applications
2. Zhang, Xing & Tian, Jie & Deng, Kexin & Wu, Yongfang & Li, Xiuli. (2010). Automatic Liver Segmentation Using a Statistical Shape Model With Optimal Surface Detection.
IEEE transactions on bio-medical engineering. 57. 2622-6. 10.1109/TBME.2010.2056369.
3. Hsu, WT, Yeh, JR, Chang, YC, et al. A computer aided for image processing of computed tomography in hepatocellular carcinoma. In: 2011 IEEE international conference on bioinformatics and biomedicine workshops, BIBMW 2011, Atlanta, GA, 12–15 November 2011, pp.942–944. New York: IEEE.
4. Seletchi, ED, Duliu, OG. Image processing and data analysis in computed tomography.
Rom J Phys 2007; 52: 667–675.
5. Chen, EL, Chung, PC, Chen, CL, et al. An automatic diagnostic system for CT liver image classification. IEEE Trans Biomed Eng 1998; 45(6): 783–794.
6. Gletsos, M, Mougiakakou, SG, Matsopoulos, GK, et al. Classification of hepatic lesions from CT images using texture features and neural networks. In: 2001 conference proceedings of the 23rd annual international conference of the IEEE engineering in medicine and biology society, Istanbul, Turkey, 25–28 October 2001, pp.2748–2751. New York: IEEE.
7. Valavanis, IK, Mougiakakou, SG, Nikita, A, et al. Evaluation of texture features in hepatic tissue characterization from non-enhanced CT images. In: Annual international conference of the IEEE engineering in medicine and biology - proceedings, Lyon, France, 2007.
pp.3741–3744. New York: IEEE.
8. Ye, J, Sun, Y, Wang, S, et al. Multi-phase CT image based hepatic lesion diagnosis by SVM. In: 2009 2nd international conference on biomedical engineering and informatics, Tianjin, China, 17 October 2009, pp.1–5. New York: IEEE.
9. Zhang, X, Furukawa, T, Zhou, X, et al. Detection of metastatic liver tumor in multi-phase CT images by using a spherical gray-level differentiation searching filter. In: Medical imaging 2011: Computer-aided diagnosis, Lake Buena Vista, FL, 25 April 2011, p.79632K. Bellingham, Washington USA: International Society for Optics and Photonics.
10. Gunasundari, S, Ananthi, MS. Comparison and evaluation of methods for liver tumor classification from CT datasets. Int J Comput Appl 2012; 39(18): 46–51.
11. Anter AM, Azar AT, Hassanien AE, El-Bendary N, Elsoud MA. Automatic computer aided segmentation for liver and hepatic lesions using hybrid segmentations techniques.
Federated Conference on Computer Science and Information Systems; New York: IEEE;
2013. pp. 193–198.
12. R.Rajagopal1, P.Subbiah, "Computer Aided Detection of Liver Tumor using SVM Classifier", International Journal of Advanced Research in Electrical,Electronics and Instrumentation Engineering (An ISO 3297: 2007 Certified Organization) Vol. 3, Issue 6, June 2014
13. Mehrdad Moghbel,*,1 Syamsiah Mashohor,1 Rozi Mahmud,2 and M. Iqbal Bin Saripan1,
"Automatic liver segmentation on Computed Tomography using random walkers for treatment planning", EXCLI J. 2016; 15: 500–517. Published online 2016 Aug 10. doi:
14. Frank Heckel*, Hans Meine, Jan H. Moltz et.al. "Segmentation-Based Partial Volume Correction for Volume Estimation of Solid Lesions in CT", IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 33, NO. 2, FEBRUARY 2014
15. Hussein Alahmera,*, Amr Ahmeda, "Computer-Aided Classification of Liver Lesions from CT Images Based on Multiple ROI", International Conference On Medical Imaging Understanding and Analysis 2016, MIUA 2016, 6-8 July 2016, Loughborough, UK
16. Y. Rakesh Kumara*, N.Moorthy Muthukrishnana, Abhishek Mahajan, "Statistical Parameter-based Automatic Liver Tumor Segmentation from Abdominal CT Scans: A Potiential Radiomic Signature", 6th International Conference On Advances In Computing
& Communications, ICACC 2016, 6-8 September 2016, Cochin, India
17. R. Vivanti1 · A. Szeskin1 · N. Lev-Cohain2 · J. Sosna2 · L. Joskowicz1, "Automatic detection of new tumors and tumor burden evaluation in longitudinal liver CT scan studies", Int J CARS, 2017
18. Marc Aubreville, Christian Knipfer2,3, Nicolai Oetter et.al., "Automatic Classification of Cancerous Tissue in Laserendomicroscopy Images of the Oral Cavity using Deep Learning", Scientic Reports 2017
19. T.S. Avsar1, S. Arica2, "Automatic Segmentation of Computed Tomography Images of Liver Using Watershed and Thresholding Algorithms", © Springer N ature Singapore Pte L td. 2018.
20. afargholi Rangraz, E., Coudyzer, W., Maleux, G. et al. Multi-modal image analysis for semi-automatic segmentation of the total liver and liver arterial perfusion territories for radioembolization. EJNMMI Res 9, 19 (2019). https://doi.org/10.1186/s13550-019-0485- x.
21. Das, A., Das, P., Panda, S.S. et al. Detection of Liver Cancer Using Modified Fuzzy Clustering and Decision Tree Classifier in CT Images. Pattern Recognit. Image Anal. 29, 201–211 (2019). https://doi.org/10.1134/S105466181902005.
22. Krishan A, Mittal D. Ensembled liver cancer detection and classification using CT images.
Proceedings of the Institution of Mechanical Engineers, Part H: Journal of Engineering in Medicine. 2021;235(2):232-244. doi:10.1177/0954411920971888
23. Otsu, N. (1979). A Threshold Selection Method from Gray-Level Histograms. IEEE Transactions on Systems, Man, and Cybernetics, 9(1), 62–66.
doi:10.1109/tsmc.1979.4310076 24. https://www.kaggle.com/
26. Mahajan, H.B., Badarla, A. & Junnarkar, A.A. (2020). CL-IoT: cross-layer Internet of Things protocol for intelligent manufacturing of smart farming. J Ambient Intell Human Comput. https://doi.org/10.1007/s12652-020-02502-0.
27. Mahajan, H.B., & Badarla, A. (2018). Application of Internet of Things for Smart Precision Farming: Solutions and Challenges. International Journal of Advanced Science and Technology, Vol. Dec. 2018, PP. 37-45.
28. Mahajan, H.B., & Badarla, A. (2019). Experimental Analysis of Recent Clustering Algorithms for Wireless Sensor Network: Application of IoT based Smart Precision Farming. Jour of Adv Research in Dynamical & Control Systems, Vol. 11, No. 9.
29. Mahajan, H.B., & Badarla, A. (2020). Detecting HTTP Vulnerabilities in IoT-based Precision Farming Connected with Cloud Environment using Artificial Intelligence.
International Journal of Advanced Science and Technology, Vol. 29, No. 3, pp. 214 - 226.
30. Mikhail, A., Kamil, I. A., & Mahajan, H. (2017). Increasing SCADA System Availability by Fault Tolerance Techniques. 2017 International Conference on Computing,
Communication, Control and Automation (ICCUBEA).
31. Mikhail, A., Kareem, H. H., & Mahajan, H. (2017). Fault Tolerance to Balance for Messaging Layers in Communication Society. 2017 International Conference on Computing, Communication, Control and Automation (ICCUBEA).