IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 13, No. 3, September 2024, pp. 2862∼2869

ISSN: 2252-8938, DOI: 10.11591/ijai.v13.i3.pp2862-2869 ❒ 2862

Automated detection of kidney masses lesions using a deep

learning approach
Ghayth AlMahadin1 , Hamza Abu Owida2 , Jamal I. Al-Nabulsi2 , Nidal Turab1 , Nour Al Hawamdeh2
1 Department of Networks and Cyber Security, Faculty of Information Technology, Al-Ahliyya Amman University, Amman, Jordan
2 Department of Medical Engineering, Faculty of Engineering, Al-Ahliyya Amman University, Amman, Jordan

Article Info ABSTRACT

Article history: Deep learning has emerged as a potent tool for various tasks, such as image
classification. However, in the medical domain, there exists a scarcity of data,
Received Oct 30, 2023
which poses a challenge in obtaining a well-balanced and high-quality dataset.
Revised Feb 19, 2024 Commonly seen issues in the realm of renal health include conditions such as
Accepted Feb 28, 2024 kidney stones, cysts, and tumors. This study is centered on the examination of
deep learning models for the purpose of classifying renal computed tomography
Keywords: (CT)-scan pictures. State-of-the-art classification models, such as convolutional
neural network (CNN) approaches, are employed to boost model performance
Convolutional neural network and improve accuracy. The algorithm is comprised of six convolutional layers
Deep learning that progressively increase in complexity. Every layer in the network utilizes a
Kidney disease uniform 3x3 kernel size and applies the rectified linear unit (ReLU) activation
Kidney masses lesions function. This is followed by a max-pooling layer that downsamples the feature
Medical image processing maps using a 2x2 pool size. Following this, a flatten layer was implemented in
order to preprocess the data for the fully linked layers. The consistent utiliza-
tion of uniform kernel sizes and activation functions throughout all layers of the
model facilitated the smooth extraction of complex features, thereby enhancing
the model’s ability to accurately identify different kidney conditions. As a result,
we achieved a high accuracy rate of 99.8%, precision is 99.8%, and F1 score of
approximately 99.7%.

This is an open access article under the CC BY-SA license.

Corresponding Author:
Ghayth AlMahadin
Department of Networks and Cyber Security, Faculty of Information Technology
Al-Ahliyya Amman University
Amman, Jordan
Email: [email protected]

1. INTRODUCTION
The human body is equipped with a pair of kidneys, which are organ structures resembling beans and
located bilaterally along the spinal column [1]. The kidney assumes a vital function in the regulation of body
fluid and solute equilibrium by means of metabolic waste secretion and filtration. Additionally, it is responsible
for the secretion of a number of hormones and contributes to the regulation of blood pressure [2].
One of the primary pathological conditions affecting the kidney is the presence of kidney masses,
which are characterised by the development of aberrant growths within the renal organ [3]. While a consider-
able proportion of renal tumours are found to be non-malignant, a substantial subset necessitates additional
medical intervention. The anatomical features, clinical manifestations, and physical examination findings
pertaining to the retroperitoneum are inadequate in terms of their ability to accurately identify renal masses.

Journal homepage: http://ijai.iaescore.com

Int J Artif Intell ISSN: 2252-8938 ❒ 2863

Various imaging modalities, including computed tomography (CT), magnetic resonance imaging (MRI), and
ultrasound, are employed in the diagnostic process of identifying masses [4], [5]. Renal masses can be cate-
gorised as either solid or cystic. Solid renal masses are of particular concern due to the potential presence of
cancer. Conversely, renal masses diagnosed as kidney stones or renal cysts are generally considered to be less
threatening [6], [7].
The development of a kidney tumor occurs when renal cells undergo a cessation of normal function
and subsequently undergo rapid proliferation. Renal cell carcinoma, originating from renal cells, exhibits
varying degrees of metastatic potential, ranging from indolent to very aggressive. Renal cell carcinoma often
presents as a solitary neoplasm. Nevertheless, it is possible for any kidney to acquire various types of cancers
[8], [9]. The incidence of kidney tumours is a significant global issue, with an annual registration of over
400,000 cases in the field. Among those, in terms of cancer prevalence, it ranks as the 9th most frequently
diagnosed cancer in males and the 14th most frequently diagnosed cancer in females [4], [10].
The necessity for an automated system to reliably classify kidney cancer histopathology images is
supported by the scarcity of available medical specialists, the possibility of interobserver variability, and the
time-consuming procedure of kidney cancer diagnosis. Such a system might enhance the speed and accuracy
of cancer diagnosis, resulting in early detection and better patient outcomes by utilising cutting-edge computer
algorithms and deep learning approaches. The adoption of any artificial intelligence (AI)-based medical system
in clinical practise, however, must first undergo extensive validation and testing to guarantee its safety and
efficacy. This problem has been addressed successfully in the past [11]-[13].
In recent years, significant advancements in the field of AI and machine learning, particularly in the
area of deep learning, have presented novel opportunities in the healthcare sector pertaining to the domains
of detection and prediction. Neural networks, which are fundamental components of deep learning method-
ologies, have exhibited remarkable efficacy in tasks such as image recognition, feature extraction, and classi-
fication [14], [15]. The algorithms and models currently available possess remarkable capabilities, providing
healthcare professionals with robust tools for analysing extensive medical data with exceptional precision and
speed. Consequently, these technologies have the potential to enhance the effectiveness of medical prescrip-
tions. Nevertheless, there are some areas that have not yet been extensively explored in the realm of AI, and
one such domain is kidney masses disorders. To date, no prior research utilising AI has been conducted on this
particular disease.
Numerous prior investigations have been undertaken to implement AI in order to automate the detec-
tion and classification of kidney medical images for kidney masses disorders with different features (normal,
cyst, stone, and tumor). Attia et al. [16] introduced a computer-aided approach designed to automatically clas-
sify ultrasound kidney disorders. The study included the examination of five distinct types of images, namely
normal, cyst, stone, tumour, and failure. The region of interest in each image was subjected to feature extrac-
tion, resulting in two sets of features: statistical features and multi-scale wavelet-based features. To minimise
the dimensionality of the feature space, principal component analysis was conducted. The chosen attributes
were employed in the development and training of a neural network classifier. The utilisation of multi-scale
wavelet-based features has resulted in a classification rate of 97% that is deemed accurate [16].
Mangayarkarasi and Jamal [17] proposed a computer-assisted tool for the processing and analysis
of ultrasound kidney images. The purpose of this tool is to classify renal pathologies based on the obtained
results. The classification of ultrasound kidney images encompasses four distinct categories, namely normal,
cyst, calculi, and tumour. The classification process involves utilising inputs from a urologist’s perspective,
which include scanned kidney ultrasound images and knowledge related to common pathologies. The images
undergo pre-processing to eliminate speckle noise with the application of median and Gaussian filters. The
utilisation of the optimal thresholding segmentation algorithm is employed to acquire the region of interest. A
collection of statistical features of the first order is extracted. The aforementioned features are utilised as inputs
during the training and testing phases of the probabilistic neural network classifier. The holdout approach is
employed, wherein 50% of the images are allocated for training purposes, while the remaining 50% are reserved
for testing. The evaluation of the classifier’s efficiency is ultimately conducted. A classification accuracy of
93.5% is achieved [17].
Wu and Yi [18] offer an automated architecture for the detection of different kidney problems. This
design utilises convolutional neural networks (CNN) and operates on abdominal ultrasound images. The model
comprises three components. The first component is responsible for selecting suitable ultrasound images of
kidneys. The second component utilises a detection model to identify the specific region occupied by the

Automated detection of kidney masses lesions using a deep learning approach (Ghayth AlMahadin)
2864 ❒ ISSN: 2252-8938

kidney within the given image. The third component employs the multifeature fusion neural network (MF-
Net) to differentiate between normal and abnormal kidneys. This is achieved by analysing the presence of
abnormal symptoms in the images. The performance of the detection model is enhanced by integrating it
with a weighted ensemble technique. The MF-Net incorporates a multi-feature fusion layer to effectively
extract distinguishing features from numerous perspectives of pictures. The three components collaborate in
a synergistic manner to autonomously identify anomalies related to renal function. A comprehensive dataset
of 3,722 abdominal ultrasound pictures, accompanied by classification and localization annotations, has been
curated for the purpose of training and evaluating the model. The experimental findings indicate that the
ensemble detection model, as presented, exhibits superior performance with an average true positive fraction
of 98.0%. Additionally, the MF-Net demonstrates an average classification accuracy of 94.67% [18].
Kanauchi et al. [19] propose the development of a deep learning model with the capability to au-
tonomously identify renal cysts in ultrasound images. Additionally, the model is designed to predict the precise
location of a pair of significant anatomical landmarks, which is crucial for accurately measuring the size of
the cysts. The detection of renal cysts was performed using a deep learning model that utilised a fine-tuned
version of YOLOv5. Additionally, the prediction of saliency maps, which indicate the location of prominent
landmarks, was carried out using a fine-tuned version of UNet++. The ultrasound images were utilised as input
for the YOLOv5 model. Subsequently, the images were cropped within the bounding box and then recognised
from the input image by the YOLOv5 model. These detected images were then fed as input to the UNet++
model and the accuracy of 86% is achieved [19].
Affan and Hussain [20] introduced a set of deep learning models designed for the purpose of classi-
fying renal CT-scan pictures. State-of-the-art classification models such as CNNs are commonly employed in
order to boost model performance and improve accuracy. Techniques such as data augmentation, normalisa-
tion, and transfer learning are utilised to achieve these objectives. The CNN architecture employed in this study
is VGG-16, which is further enhanced by the development of a bespoke model for further training. In order to
get a more balanced distribution of the dataset, data augmentation techniques were employed. As a result, the
accuracy of the model reached 98%, with an F1 score of about 99% [20].
Even though earlier studies have conscientiously explored the utilization of AI in the identification
and categorization of kidney masses, a conspicuous void remains as they have yet to explicitly tackle the
exhaustive classification criteria via the analysis of CT images. This study aims to employ deep learning
techniques, specifically neural networks, to facilitate the efficient and accurate diagnosis of kidney masses
disorders. The suggested architecture demonstrates significant efficacy in accurately identifying kidney images,
precisely localising kidneys within the images, and effectively recognising various kidney disorders.

2. METHOD
2.1. Dataset
The dataset was collected from various hospitals in Dhaka, Bangladesh, utilizing the picture archiving
and communication system (PACS). The patients included in the dataset had previously received diagnoses of
kidney tumors, cysts, normal conditions, or the presence of kidney stones. Both the coronal and axial sections
were chosen from both contrast-enhanced and non-contrast investigations following a technique for imaging
the whole abdomen and urogram. The Dicom study was meticulously chosen, with a focus on one diagnosis at
a time. From these selections, a collection of Dicom photos pertaining to the specific region of interest for each
radiological finding was generated. Subsequently, the patient-specific data and meta data were removed from
the Dicom photos, and the Dicom images were then transformed to a lossless jpg image format. Following
the conversion process, each picture discovery underwent subsequent verification by both a radiologist and a
medical technologist in order to reaffirm the accuracy of the data. The dataset that was generated consists of
12,446 distinct data points, categorized as follows: 3,709 for cyst, 5,077 for normal, 1,377 for stone, and 2,283
for tumor [21].
2.2. Artificial intelligence-based detection methods
2.2.1. Preprocessing phase
The preprocessing phase plays a pivotal role in preparing the data for the model. In our kidney disease
classification project, we conducted several important preprocessing steps to ensure the data was suitable for
feeding into our neural network. First, the dataset was organized using a ratio-based split using the splitfolders
library to create training, validation, and test sets. This resulted in a stratified division of the data into 90% for

Int J Artif Intell, Vol. 13, No. 3, September 2024: 2862–2869

Int J Artif Intell ISSN: 2252-8938 ❒ 2865

training, 5% for validation, and 5% for testing. This approach facilitated the creation of distinct subsets for
training, validation, and testing. Importantly, the split was stratified to maintain class balance across subsets,
thus mitigating the risk of bias during model training.
The data was then preprocessed for the model. Images were resized to a consistent resolution of
200x200 pixels to ensure uniformity in the input data and reduce computational complexity [22]. Additionally,
images were converted to grayscale to reduce the computational load. The images were organized into batches
for efficient training using the ImageDataGenerator from the Keras library. These batches were configured
for rescaling pixel values to the range [0, 1] [23]. These batches were not only instrumental in efficient data
handling during training but also facilitated the implementation of data augmentation techniques such as rota-
tion, scaling, and flipping. By augmenting the training data, we augmented the diversity of the dataset, thus
enhancing the model’s ability to generalize to unseen examples.
2.2.2. Model architecture
The model architecture employed for the kidney disease classification entails a sequential CNN struc-
ture [24]. The network consists of six convolutional layers sequentially increasing in complexity, with 32, 32,
64, 64, 128, and 128 filters, respectively. Each layer employs a consistent 3x3 kernel size and rectified linear
unit (ReLU) activation function. Using 3x3 filters in the CNN architecture enables efficient hierarchical feature
extraction, maintaining spatial information, and facilitating translation invariance for effective pattern recogni-
tion with parameter efficiency [25]. The filters are followed by a max-pooling layer utilizing a 2x2 pool size to
downsample the feature maps [26]. This downsampling process aids in reducing the spatial dimensions of the
feature maps while retaining essential information. Consequently, computational overhead is minimized, and
the network’s robustness is enhanced. Subsequently, a flattened layer was introduced to transform the multi-
dimensional feature maps into a one-dimensional array. This step prepares the data for input into the subsequent
dense layers (fully connected layers). The network also integrated a dense layer with 512 units, maintaining the
ReLU activation function, and culminated in an output layer with four units, employing softmax activation to
compute the probability distribution across classes, thereby enabling multi-class classification with confidence
scores [27]. This uniformity in the kernel size and activation function across all layers facilitated the seamless
extraction of intricate features, contributing to the model’s robust performance in accurately identifying various
kidney conditions.

3. RESULTS AND DISCUSSION

Over the course of five training epochs, the model demonstrated notable performance improvements.
The accuracy of the training and validation dataset is shown in Figure 1. The accuracy of the training dataset
rose significantly, starting at 52.83% and reaching an impressive 97.82% by the final epoch. In addition, the
validation dataset validated the model’s robustness and generalizability, achieving an exceptional accuracy of
99.92%. The precision scores of training and validation datasets shown in Figure 2 experienced a substantial
increase, peaking at 97.93% in the training dataset and 99.92% in the validation dataset, underlining the model’s
capability for precise classification. Meanwhile, the recall values remained consistently high as shown in
Figure 3, confirming the model’s competence in identifying relevant instances of kidney diseases and achieving
a recall of 97.71% at the end of training and in the validation dataset 99.92%.

Figure 1. Training and validation accuracy Figure 2. Training and validation precision

Automated detection of kidney masses lesions using a deep learning approach (Ghayth AlMahadin)
2866 ❒ ISSN: 2252-8938

Figure 3. Training and validation recall

The test dataset results reinforced the model’s efficacy, showing an accuracy of 99.76%, precision
exceeding 99.77%, recall surpassing 99.66%, and an F1 score of 99.72%. The confusion matrix as shown in
Figure 4 offered a comprehensive view of the model’s classification performance across various kidney con-
ditions, highlighting its accuracy in correctly identifying the classes with minimal misclassifications. These
outcomes emphasize CNN’s proficiency in accurately diagnosing kidney diseases, suggesting its potential ap-
plication in clinical settings for precise and timely interventions.

Figure 4. The confusion matrix

The receiver operating characteristic (ROC) curves in Figure 5 provide additional evidence of the
model’s classification performance. The macro-average ROC curve exhibited an area under the curve (AUC)
of 1.0, indicating the model’s remarkable ability to discriminate between different kidney conditions with preci-
sion. The model exhibited an exceptionally elevated capacity for the classification of kidney disorders, yielding
a substantial abundance of true positives coupled with a minimal occurrence of false negatives. Moreover, the
model demonstrated a commendable level of accuracy and precision, characterized by a minimal occurrence of

Int J Artif Intell, Vol. 13, No. 3, September 2024: 2862–2869

Int J Artif Intell ISSN: 2252-8938 ❒ 2867

false positives. These findings offer valuable insights into the robust performance of the model, particularly in
its ability to accurately calcify kidney disorders.

Figure 5. The ROC curves and AUC

Our study suggests that the high level of accuracy and precision observed in the classification of kid-
ney disorders is intricately linked with outstanding performance in meeting kidney disorders treatment criteria.
The proposed method stands to gain substantial advantages from the perspective of medical experts, as it offers
a robust framework that aligns with their viewpoints, all the while ensuring minimal disruption to patients’ ther-
apy plans. Table 1 presents a comparison between our proposed kidney disorders detection and classification
models and the existing approaches, utilising accuracy as the parameter of evaluation.

Table 1. A comparison between our proposed kidney disorders detection and classification models and the
existing approaches
Reference Technique Accuracy (%)
[16] CNN 97
[17] PNN-based model 93.5
[18] MF-Net 94.7
[19] the UNet++ model 86
[20] VGG-16 99
Proposed study EfficientNetB3 100

4. CONCLUSION
Deep learning has developed as a powerful technology that may be utilized for a variety of tasks,
including the classification of images. However, there is a lack of data in the medical field, which makes
it difficult to acquire a dataset that is both well-balanced and of a high quality. In order to effectively train
algorithms and improve their performance, deep learning models are extremely reliant on having access to a
considerable amount of data. Conditions such as kidney stones, cysts, and tumors are examples of problems
that are frequently seen in the field of renal health. The primary focus of this research is on the investigation
of deep learning models for the purpose of image classification using renal CT scans. It is common practice
to apply cutting-edge classification models, such as those based on CNN techniques, in order to increase the
accuracy and performance of models. The network is made up of six convolutional layers, each of which is
more complicated than the one that came before it. These layers each have a filter size of 32, followed by 64,
then 128 and finally 128, in that order. Each layer of the network makes use of the ReLU activation function and
employs a kernel size that is consistently 3 by 3, making it uniform throughout. After this comes a max-pooling
layer, which uses a pool size of 2x2 to perform a downsampling operation on the feature maps. After this, a
flatten layer was built in order to preprocess the data for the completely linked layers. This was done after the
previous step. In addition to this, the network had a densely connected layer that was composed of 512 units
and activated using the ReLU. After this, there was a four-unit output layer that used the softmax activation

Automated detection of kidney masses lesions using a deep learning approach (Ghayth AlMahadin)
2868 ❒ ISSN: 2252-8938

function to make multi-class classification easier. This came after the previous step. The smooth extraction
of intricate features was made possible by the model’s consistent use of regular kernel sizes and activation
functions across all of its layers. This helped the model get better at properly identifying a variety of kidney
diseases. As a consequence of this, we were able to accomplish a high accuracy rate of 99.8%, with precision at
the same level, and an F1 score of roughly 99.7%. In the future, there is anticipation for the detection of precise
lesions through the utilization of segmentation techniques. This approach can aid in identifying the specific
location of damage within the kidney. The precise location of the tumor, cyst, or stone within the kidney is
unknown. This tool has the potential to aid physicians in making prognoses.

REFERENCES
[1] A. Baranski, “Basic anatomy of the kidney, ureters and the urinary bladder, and their functions,” Kidney Transplantation, pp. 1–32,
2023, doi: 10.1007/978-3-030-75886-8 1.
[2] M. A. Wallace, “Anatomy and physiology of the kidney,” AORN Journal, vol. 68, no. 5, pp. 799–820, 1998, doi: 10.1016/S0001-
2092(06)62377-6.
[3] C. Bergmann, L. M. G. -Woodford, P. C. Harris, S. Horie, D. J. M. Peters, and V. E. Torres, “Polycystic kidney disease,” Nature
Reviews Disease Primers, vol. 4, no. 1, pp. 1–24, 2018, doi: 10.1038/s41572-018-0047-y.
[4] Y. Kamkova, H. Ali Qadir, O. Jakob, and R. P. Kumar, “Kidney and tumor segmentation using combined deep learning method,” in
Submissions to the 2019 Kidney Tumor Segmentation Challenge: KiTS19, 2019, pp. 1–11, doi: 10.24926/548719.091.
[5] R. Sharma and N. Choubey, “Kidney tumor segmentation: a review,” AIP Conference Proceedings, vol. 2782, no. 1, 2023, doi:
10.1063/5.0154320.
[6] L. I. Vázquez et al., “Cystic renal lesions: a systematic review of diagnosis and treatment,” Kidney Cancer, vol. 7, no. 1, pp. 95–114,
2023, doi: 10.3233/KCA-230002.
[7] S. R. Khan et al., “Kidney stones,” Nature Reviews Disease Primers, vol. 2, no. 1, 2016, doi: 10.1038/nrdp.2016.8.
[8] J. Docs, G. Kovacs, and L. Peterfi, “End-stage kidney disease: a never healing wound leading to another never healing wound, renal
cancer,” Journal of Nephrology, vol. 36, no. 6, pp. 1673–1681, 2023, doi: 10.1007/s40620-023-01694-w.
[9] J. J. Hsieh et al., “Renal cell carcinoma,” Nature Reviews Disease Primers, vol. 3, no. 1, 2017, doi: 10.1038/nrdp.2017.9.
[10] WCRF, “Kidney cancer statistics,” World Cancer Research Fund. 2022. [Online]. Available: https://www.wcrf.org/cancer-
trends/worldwide-cancer-data/
[11] S. Reddy, “Explainability and artificial intelligence in medicine,” The Lancet Digital Health, vol. 4, no. 4, pp. e214–e215, 2022, doi:
10.1016/S2589-7500(22)00029-2.
[12] F. S. Nabiyeva, S. S. Umarova, and S. I. Umirkulova, “Artificial intelligence in medicine,” Journal of New Century Innovations, vol.
30, no. 3, pp. 153–155, 2023.
[13] Z. Xiao et al., “Emerging early diagnostic methods for acute kidney injury,” Theranostics, vol. 12, no. 6, pp. 2963–2986, 2022, doi:
10.7150/thno.71064.
[14] F. Yousefirizi, P. Decazes, A. Amyar, S. Ruan, B. Saboury, and A. Rahmim, “AI-based detection, classification and predic-
tion/prognosis in medical imaging,” PET Clinics, vol. 17, no. 1, pp. 183–212, 2022, doi: 10.1016/j.cpet.2021.09.010.
[15] C. Liu, D. Jiao, and Z. Liu, “Artificial intelligence (AI)-aided disease prediction,” BIO Integration, vol. 1, no. 3, pp. 130–136, 2020,
doi: 10.15212/bioi-2020-0017.
[16] M. W. Attia, F. E. Z. A. -Chadi, H. E. -Din, and N. Mekky, “Classification of ultrasound kidney images using PCA
and neural networks,” International Journal of Advanced Computer Science and Applications, vol. 6, no. 4, 2015, doi:
10.14569/IJACSA.2015.060407.
[17] T. Mangayarkarasi and D. N. Jamal, “PNN-based analysis system to classify renal pathologies in kidney ultrasound Images,” in 2017
2nd International Conference on Computing and Communications Technologies (ICCCT), 2017, pp. 123–126, doi: 10.1109/IC-
CCT2.2017.7972258.
[18] Y. Wu and Z. Yi, “Automated detection of kidney abnormalities using multi-feature fusion convolutional neural networks,”
Knowledge-Based Systems, vol. 200, 2020, doi: 10.1016/j.knosys.2020.105873.
[19] Y. Kanauchi et al., “Automatic detection and measurement of renal cysts in ultrasound images: a deep learning approach,” Health-
care, vol. 11, no. 4, 2023, doi: 10.3390/healthcare11040484.
[20] A. Affan and S. Hussain, “Detecting multi-class kidney abnormalities using Deep learning,” in 2023 17th International
Conference on Ubiquitous Information Management and Communication (IMCOM), 2023, pp. 1–7, doi: 10.1109/IM-
COM56909.2023.10035666.
[21] M. N. Islam, M. Hasan, M. K. Hossain, M. G. R. Alam, M. Z. Uddin, and A. Soylu, “Vision transformer and explainable transfer
learning models for auto detection of kidney cyst, stone and tumor from CT-radiography,” Scientific Reports, vol. 12, no. 1, 2022,
doi: 10.1038/s41598-022-15634-4.
[22] A. Géron, Hands-on machine learning with scikit-learn, keras, and tensorFlow. Sebastopol, USA: O’Reilly Media, 2023.
[23] F. Chollet, Deep learning with python. Shelter Island, New York: Manning Publications, 2021.
[24] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet classification with deep convolutional neural networks,” Communications
of the ACM, vol. 60, no. 6, pp. 84–90, 2017, doi: 10.1145/3065386.
[25] S. Karen and Z. Andrew, “Very deep convolutional networks for large-scale image recognition,” in 3rd International Conference on
Learning Representations, ICLR 2015 - Conference Track Proceedings, 2015, p. 14.
[26] S. Liu, Z. Song, G. Liu, C. Xu, H. Lu, and S. Yan, “Street-to-shop: cross-scenario clothing retrieval via parts align-
ment and auxiliary set,” in 2012 IEEE Conference on Computer Vision and Pattern Recognition, 2012, pp. 3330–3337, doi:
10.1109/CVPR.2012.6248071.
[27] I. Goodfellow, Y. Bengio, and A. Courville, Deep learning. Cambridge, Massachusetts: MIT Press, 2016.

Int J Artif Intell, Vol. 13, No. 3, September 2024: 2862–2869

Int J Artif Intell ISSN: 2252-8938 ❒ 2869

BIOGRAPHIES OF AUTHORS

Ghayth AlMahadin holds a Ph.D. in Artificial Intelligence-Machine Learning from Not-

tingham Trent University, UK in 2022. He also received his B.Sc. in Computer Engineering from
Mutah University, Jordan in 2005 and M.Sc. in Cloud Computing and Machine Learning from Not-
tingham Trent University, UK in 2017. He is also an assistant professor at Al-Ahliyya Amman
University. His research interests include wearable technologies and machine learning applications
in different fields. He can be contacted at email: [email protected].

Hamza Abu Owida has completed his Ph.D. from Keele university, UK. He was a
postdoctoral Research Associate: Developing xeno-free nanofibrous scaffold methodology for hu-
man pluripotent stem cell expansion, differentiation and implantation towards a therapeutic prod-
uct, Keele University, Institute for Science and Technology in Medicine (ISTM), Staffordshire /UK.
He is assistant professor in Department of Medical Engineering at Al-Ahliyya Amman Univer-
sity. He has published more than 30 papers in reputed journals. He can be contacted at email:
[email protected].

Jamal I. Al Nabulsi received his Ph.D. degree from Saint Petersburg Electro Technical
University, Saint Petersburg, Russia, in 1999. He joined the teaching staff of the Department of
Biomedical Engineering, Faculty of Engineering, The Hashemite University, Zarqa, Jordan from
2002 to 2013. In 2013, he moved to Al-Ahliyya Amman University, and he is currently a Professor at
the Medical Engineering Department there. He published more than 40 scientific papers in journals
and international conference proceedings. His research interests are biomedical sensors, generating
and harvesting energy from human motion digital signal processing, and image processing. He can
be contacted by this email: [email protected].

Nidal Turab holds Ph.D. in computer science and he is also a Professor at the Depart-
ment of Networks and Cyber Security, Al-Ahliyya Amman University, Jordan. His research interests
include WLAN security, computer networks security and cloud computing security, eLearning and
internet of things. He can be contacted by this email: [email protected].

Nour Al Hawamdeh has completed her bachelor degree from Department of Medical
Engineering in Al-Ahliyya Amman University. Her final year project was in application of artifi-
cial intelligence for kidney abnormalites detection and classificatin. She can be contacted at email:
[email protected].

Automated detection of kidney masses lesions using a deep learning approach (Ghayth AlMahadin)