Page 271 - 2024-Vol20-Issue2

267 |
lung cancer detection techniques. As computer technology ad-
vances swiftly, the relevance of machine learning techniques
in lung cancer diagnosis is becoming more pronounced. The
exploration of diverse algorithms for lung cancer prediction is
ongoing among researchers. However, studies on lung cancer
prediction indicate certain limitations in these methods. For
instance, selecting an appropriate kernel function for SVM
poses a challenge. In the case of Naive Bayes, the requirement
for a known prior probability can be a drawback; inaccuracies
in the assumed prior distribution may result in suboptimal
prediction outcomes [8]. Lung cancer detection systems and
research must be improved in terms of accuracy in order to
work properly. The development of hybrid methodologies can
enhance electronic diagnosis, early-stage lung cancer detec-
tion, localization detection, tumor segmentation, and stage
classification. We apply these cutting-edge techniques inside
a hybrid model to use in the healthcare sector since their effec- For the purpose of assisting radiologists, Masood et al. de-
tiveness inspires us. In this effort, we solved these problems
of the existing scenario and created a new prototype named
”Hybrid Lung Cancer Stage Classifier and Diagnosis Model”
(Hybrid-LCSCDM) framework; this uses a feature extraction
stage based on the VGG16 model and a classification stage
based on the XGBoost algorithm to identify lung cancer as
early as possible.
A. Main Contributions
We suggested an intelligent diagnosis module named (Hybrid-
LCSCDM) to identify and classify lung cancers at an early
stage. The proposed model was employed based on a hybrid
approach for both stage classification and feature extraction,
which reduced detection time complexity and increased detec-
tion rate accuracy.
1. A hybrid system is presented for impactful analysis.
2. This model’s dependence is between image-quantitative
and clinical features; for a more precise computer-aided
diagnosis, it also offers a novel feature representation.
3. This technique employs VGG16 for feature extraction
and an XGBoost classifier for lung cancer classification
and diagnosis in an efficient manner.
4. The Lung Cancer Dataset (IQ-OTH/NCCD) was used
to assist us in carrying out the tests in this study.
5. The suggested Hybrid-LCSCDM has been used to test
the effectiveness of a novel technique also presented in
this research.
The organization of the paper is as follows: Section II. delves
into a comprehensive literature review, where we explore and
synthesize relevant previous. Section III. , titled ’Methodol-
ogy,’ details the methodologies and approaches we employed
Qadir, Abdalla & Abd
in our research. In Section IV. , we present our results, and
finally, Section V. is dedicated to the discussion.
II. LITERATURE REVIEW
Sharma et al. examined different kinds of colon and lung (3
malignant and two benign) tissues by creating a diagnosis
system utilizing the histopathological dataset. According to
the data, the proposed model in this research can successfully
identify malignant tissues up to 96.33 percent of the time [9].
The success of the CAD system relies on the extraction of rele-
vant texture features for distinguishing between cancerous and
noncancerous nodules, and the authors mention area, calcifi-
cation, shape, size, and contrast enhancement as features. The
appropriateness and sufficiency of these features in capturing
the variations in lung nodules need careful consideration.
vised a method for diagnosing lung nodules based on 3D-
DCNN employing computer-aided diagnosis assistance sys-
tems. In this work, a computer-aided diagnostic (CAD) model
was validated and trained on the LIDC-IDRI, ANODE09, and
LUNA16 datasets [10]. A deep CNN-based binary classifier
was developed by Kalaivani et al. through the utilization of the
DenseNet model, with the aim of identifying patients with ei-
ther benign or aggressive lung cancer [11], as the researchers
used 201 lung scans in a dataset in this study, 85% of the
dataset was used to train the model while using the remain-
ing of the images for the testing phase, test results showed
that the suggested approach achieved 90.85% accuracy on the
mentioned dataset.
To classify various cancer types using genetic data with ref-
erence to tumor RNA sequences, ElNabi et al. provided a
special optimized deep-learning model, namely a convolu-
tional neural network (CNN) and decision tree (BPS, O-DT)
while making use of the particle swarm optimization. This
study addressed the performance criteria, including recall, F1-
score, and precision [11].
Qin et al. explained how to collect, analyze, and fuse multi-
type interdependent characteristics to determine EGFR mu-
tation status utilizing computer-assisted diagnostics. In this
study, the CNN-RNN architecture is employed to provide a
novel hybrid network paradigm. Using CNN, quantitative
image characteristics are retrieved, and a model is created
to show how different types of features relate to one another
[12].
To detect lung cancer, Joshua et al. proposed an unsuper-
vised learning model for 3D CNN [13]. An improved gradient
activation function in the 3D CNN binary classifier model in-
creases the visibility of lung tumors. The suggested AlexNet
detection method employing the LUNA dataset is contrasted
with a 2D CNN learning classifier that is already in use. The
proposed model is useless since there was not insufficient data