Improving Disease Classiﬁcation Accuracy with Hybrid CNN-RNN

Architectures for Lung Tumors

Vishal R Patil, Vineet S Hiremani, Adil Mulimani, Shreeniwas R Kolagal and Channabasappa Muttal

School of Computer Science and Engineering (SoCSE), KLE Technological University, Hubballi, India

Keywords:

Lung Nodules, CNN, RNN, CT Scans, Medical Imaging, Hybrid Model.

Abstract:

The detection of lung nodules is essential in medical imaging, playing a critical role in diagnosing lung cancer

at its early stages and supporting timely treatment. This study introduces a hybrid CNN-RNN model designed

to enhance the accuracy and precision of lung nodule identiﬁcation in computed tomography (CT) scans. The

framework combines the spatial feature extraction capabilities of Convolutional Neural Networks (CNNs)

with the temporal sequence analysis strengths of Recurrent Neural Networks (RNNs), effectively integrating

spatial and temporal information for improved detection performance. Trained on a labeled dataset of CT

images, the model’s performance was assessed using metrics such as precision, recall, F1 score, and area

under the curve (AUC). The proposed method surpassed existing techniques, achieving an accuracy of 96.1%,

an F1 score of 0.8434, an AUC of 0.901, a precision of 76.02%, and a recall of 94.81%. It demonstrated

signiﬁcant advancements over hybrid CNN-LSTM models previously used in related ﬁelds like Parkinson’s

disease detection, agricultural disease analysis, and lung cancer prognosis estimation, which recorded lower

precision, recall, and F1 scores. These ﬁndings highlight the potential of CNN-RNN architectures for lung

nodule detection and their promise in advancing early lung cancer diagnosis.

1 INTRODUCTION

Lung cancer continues to be the leading cause

of cancer-related deaths worldwide, accounting for

nearly 18% of all cancer fatalities annually (Society,

2024). Early detection plays a pivotal role in im-

proving survival rates, as identifying lung nodules at

an initial stage provides the best chance for effective

treatment and favorable patient outcomes. Computed

tomography (CT) imaging has emerged as a critical

tool in identifying these nodules. However, interpret-

ing CT scans manually is both time-consuming and

subject to observer variability, often leading to incon-

sistent diagnoses (Patel and Sharma, 2024). This in-

consistency arises from the inherent complexity of an-

alyzing three-dimensional imaging data, coupled with

subtle variations in nodule size, shape, and location

(Verma and Singh, 2023). To mitigate these chal-

lenges, there is growing interest in developing auto-

mated systems that can enhance the efﬁciency, accu-

racy, and reliability of lung nodule detection while re-

ducing clinicians’ workload.

Advances in artiﬁcial intelligence (AI), partic-

ularly in deep learning, offer promising solutions

to these challenges. Convolutional neural networks

(CNNs) have transformed medical imaging by en-

abling the extraction of intricate spatial features from

CT scans, facilitating precise lung nodule identiﬁca-

tion (Lee and Gupta, 2023). These models excel at

handling large datasets and identifying patterns that

may elude human interpretation (P. Mishra and Ku-

mar, 2024). In addition, recurrent neural networks

(RNNs), including long short-term memory (LSTM)

models, have proven effective for processing sequen-

tial data and capturing temporal relationships, further

enhancing diagnostic potential (Verma and Kumar,

2023). Combining CNNs for spatial analysis with

RNNs for temporal modeling has led to signiﬁcant

progress in lung cancer detection and classiﬁcation

(Kumar and Sharma, 2024). This hybrid approach is

particularly valuable in scenarios involving serial CT

imaging, where tracking changes in nodule character-

istics over time is crucial for early diagnosis (et al.,

2024).

Despite the potential of hybrid CNN-RNN archi-

tectures, several hurdles must be overcome before

they can be integrated into clinical practice. One of

the most pressing challenges is the limited availabil-

ity of large, annotated datasets, which are essential for

training robust AI models. Generating these datasets

574

Patil, V. R., Hiremani, V. S., Mulimani, A., Kolagal, S. R. and Muttal, C.

Improving Disease Classiﬁcation Accuracy with Hybrid CNN-RNN Architectures for Lung Tumors.

DOI: 10.5220/0013597000004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 2, pages 574-580

ISBN: 978-989-758-763-4

requires expert annotation of extensive medical image

collections, a resource-intensive and time-consuming

process (Sharma and Lee, 2023). Additionally, the

computational demands of processing high-resolution

volumetric CT scans pose signiﬁcant challenges for

real-time clinical use, where timely decision-making

is critical (Liu and Zhang, 2022). Model generaliza-

tion across diverse clinical settings is further com-

plicated by variations in imaging protocols, scanner

conﬁgurations, and patient demographics (Mehta and

Agarwal, 2024). To address these issues, standardiz-

ing preprocessing methods has become a priority to

enhance the adaptability and reliability of these mod-

els across various medical environments (Rao and Pa-

tel, 2023). Moreover, ensuring that these advanced

AI systems integrate seamlessly into clinical work-

ﬂows is essential for bridging the gap between re-

search innovations and practical application (Wang

and Li, 2023).

In this proposed work, we describe the develop-

ment of a hybrid framework for the accurate and ef-

fective detection of lung nodules in CT scans that

combines Convolutional Neural Networks (CNNs)

and Recurrent Neural Networks (RNNs). The com-

plex intricacies of nodules are captured from indi-

vidual slices by the CNN component, which is ex-

cellent at extracting spatial features. In order to pro-

vide a more comprehensive understanding of nodule

features, the RNN component models temporal rela-

tionships across successive CT slices. These elements

work together to create a pipeline that tackles issues

including nodule size, shape, and location variability.

The goal of the hybrid model is to lessen frequent

challenges in medical imaging, like observer variabil-

ity and the laborious process of manual interpretation.

Our method improves lung nodule detection consis-

tency and reliability by integrating spatial and tem-

poral data processing. Additionally, the framework’s

modular design facilitates adaptability to a range of

clinical needs, including focused diagnostic activities

and extensive screenings.

In situations when early diagnosis is essential to

enhancing patient outcomes, this pipeline is very ben-

eﬁcial. It makes use of sophisticated preprocessing

methods, such as data augmentation, segmentation,

and normalization, to strengthen the system’s resis-

tance to changes in patient demographics and imaging

protocols. By using a hybrid CNN-RNN technique,

the model is guaranteed to be able to process intricate

medical data with computational efﬁciency appropri-

ate for real-time applications.

In this paper, we discuss our work in the follow-

ing sections. In Section 2, a comprehensive back-

ground study is presented, exploring recent advance-

ments in pulmonary nodule detection using hybrid

CNN-RNN techniques and other machine learning

approaches. Section 3 delves into the methodology,

detailing the dataset preparation, the hybrid CNN-

RNN model architecture, and the training process em-

ployed in our proposed approach. Section 4 highlights

the results and performance metrics of the model, in-

cluding comparisons with contemporary methods. Fi-

nally, Section 5 provides the conclusion, summarizing

the ﬁndings and outlining potential future improve-

ments for this work.

2 BACKGROUND STUDY

Recent advancements in machine learning have sig-

niﬁcantly enhanced the detection and classiﬁcation of

pulmonary nodules, which is crucial for early lung

cancer diagnosis. A variety of studies have explored

different machine learning techniques beyond deep

learning, contributing to this progress.

Marinakis, Karampidis, and Papadourakis (Mari-

nakis et al., 2024) conducted an in-depth review of

the existing literature on pulmonary nodule detection,

segmentation, and classiﬁcation through the use of

deep learning. Their analysis underscores the crit-

ical role of extracting nodule data from radiologist-

annotated pixel data to effectively train models. They

examined methods for creating 2D and 3D nodule

patches, emphasizing the beneﬁts of multi-view patch

usage in improving model outcomes. The review

also addresses challenges and outlines prospective ad-

vancements in applying deep learning to pulmonary

nodule research.

Liu et al. (Liu et al., 2023) introduced a data

augmentation framework coupled with an embedding

mechanism to enhance pulmonary nodule detection

and classiﬁcation, especially in limited-data scenar-

ios. Their methodology includes a 3D pixel-based sta-

tistical algorithm to create synthetic nodules, which

are merged with healthy lung samples to generate ex-

panded training datasets. The embedding approach

they proposed improves feature representation, lead-

ing to better accuracy and reliability across both de-

tection and classiﬁcation tasks, with potential appli-

cability to other imaging domains.

Wang et al. (Wang et al., 2022) introduced a

deep learning model speciﬁcally designed for diag-

nosing solid pulmonary nodules. This multi-task

framework not only determines lesion malignancy

but also highlights critical features, enabling inter-

pretability by visually identifying these manifesta-

tions. The model achieved an impressive test AUC

of 0.992 on the LIDC dataset and 0.923 on an inter-

Improving Disease Classiﬁcation Accuracy with Hybrid CNN-RNN Architectures for Lung Tumors

575

nal dataset. By incorporating manifestation-speciﬁc

tasks, the model enhanced malignancy classiﬁcation

accuracy, improving its utility in clinical settings and

facilitating better collaboration with radiologists.

Hesse et al. (Hesse et al., 2020) explored transfer

learning techniques to determine the origins of pri-

mary tumors in lung nodules using spectral CT im-

ages. They implemented a 3D convolutional neural

network (CNN) for nodule detection and leveraged

a pre-trained model as a feature extractor to classify

nodules as benign, primary lung cancer, or metas-

tases. This approach achieved a classiﬁcation accu-

racy of 78% in a three-class setting, demonstrating

the potential of pre-trained models to deliver robust

results with minimal ﬁne-tuning.

Chen and Xie (Chen and Xie, 2024) proposed a

novel detection network designed to handle hard sam-

ples in nodule detection. Their method integrates

deformable convolution with self-paced learning and

achieved competitive results on the LUNA16 dataset.

This approach underscores the importance of priori-

tizing difﬁcult cases to improve overall detection ac-

curacy.

Hosseini et al. (Hosseini et al., 2022) provided

a systematic review of deep learning applications for

early-stage lung cancer diagnosis, examining a vari-

ety of models and their performance. Their study

highlights ongoing challenges and proposes strategies

to reﬁne diagnostic tools, offering valuable insights

for clinicians and researchers.

Aslani et al. (Aslani et al., 2022) proposed a time-

series deep learning architecture combining multi-

modal data, such as nodule-speciﬁc, lung-speciﬁc,

and demographic details. Their approach demon-

strated superior performance in malignancy predic-

tion, showcasing the value of integrating longitudinal

data for lung cancer screening.

Al Ewaidat and El Brag (Ewaidat and El Brag,

2022) utilized a convolutional neural network-based

YOLOv5 model for localizing nodules in CT scans.

Their method achieved an accuracy of 92.27% for

nodule identiﬁcation, illustrating the effectiveness of

CNN-based solutions in medical imaging.

These studies collectively highlight the broad

scope of machine learning in advancing pulmonary

nodule detection and classiﬁcation. Techniques in-

volving clinical data integration, handling complex

samples, and applying multiscale analysis have driven

improvements in the early diagnosis of lung cancer.

3 METHODOLOGY

This section details the proposed hybrid CNN-RNN

framework for lung nodule detection, incorporating

both spatial feature extraction and temporal sequence

modeling using the LUNA 16 (Grand Challenge,

2016) dataset.

3.1 Dataset Description

The dataset used for our proposed work is LUNA16

derived from LIDC-IDRI dataset (Grand Challenge,

2016), which includes low-dose lung CT images,

which is divided into 10 subsets to provide tenfold

cross-validation. The dataset is a collection of 888

CT scans with about 1,186 lung nodules annotated by

doctors. All the CT scans are stored in .mhd format

for medical imaging with dimensions around 512 x

512 pixels with minimum voxel spacing of 1.00mm

and have a slice thickness of less than 2.5mm, which

provides higher-resolution imaging suitable for analy-

sis. The collection of lung nodules in diameter range

from 3.0 mm to 28.3 mm, with an average diameter

of 8.3 mm and position co-ordinates mentioned in a

.csv ﬁle for set of candidate nodules.

3.2 Dataset Preparation

The LUNA 16 (Grand Challenge, 2016) dataset, con-

sisting of annotated CT scans, was used to train and

evaluate the model. A preprocessing pipeline was ap-

plied to normalize, segment, and augment the data.

In normalization pixel values were scaled to the

range [0, 1] for uniformity. In segmentation Lung

regions were isolated using thresholding techniques

based on Hounsﬁeld Unit (HU) values to exclude ir-

relevant background. In resizing each CT slice was

resized to 32 ×32 ×32 voxels to optimize the compu-

tational efﬁciency of the CNN model. In augmenta-

tion random rotations, ﬂipping, and intensity shifting

were applied to simulate real-world variations, pre-

venting overﬁtting.

3.3 Model Architecture

The model consists of a CNN for feature extraction

followed by an RNN for temporal modeling 1(ﬁgure

1). The proposed architecture leverages a U-Net-

based segmentation model to extract regions of inter-

est from medical images, followed by a classiﬁcation

model to determine whether the regions are cancer-

ous. If cancerous, the system performs type classiﬁ-

cation to identify malignancy or benignity, enabling

precise diagnostic outcomes.

INCOFT 2025 - International Conference on Futuristic Technology

576

Figure 1: Pipeline of the proposed implementation showing the integration of U-Net for image segmentation and Hybrid

CNN-RNN for tumor classiﬁcation

3.3.1 CNN Component

The CNN is based on a modiﬁed U-Net architecture

designed to capture spatial features. The convolu-

tional operation is deﬁned as follows:

y = σ(W ·x +b) (1)

where x is the input, W is the convolutional kernel,

b is the bias term, and σ is the ReLU activation func-

tion. Batch normalization and dropout are applied af-

ter each convolutional layer to stabilize training and

mitigate overﬁtting. The U-Net Architecture 2 (ﬁgure

2) is used for CT image segmentation of the lungs, to

extract the nodule features present and is then passed

to the classiﬁcation model.

Figure 2: Unet architecture (Ronneberger et al., 2015)

3.3.2 RNN Component

The temporal modeling component uses an LSTM

network, which processes sequential data extracted

from adjacent CT slices.

3.3.3 Output Layer

Sigmoid function is used as activation function in our

ﬁnal output layer, producing a probability p for the

presence of a nodule:

3.4 Loss Function and Optimization

The loss function used was the binary cross-entropy

(BCE) loss function used to evaluate model perfor-

mance. The BCE loss is deﬁned as:

BCE Loss = −

∑

i=1

log(p

)+(1 −x

)log(1 −p

)

(2)

where S is the total number of samples, x

is the

true label, and p

is the predicted probability for each

sample.

Optimizer: The Adam optimizer was applied

with an initial learning rate of 10

−4

. The parameter

update rule can be expressed as:

= θ

t−1

−

η ·m

√

+ ε

(3)

where θ

represents the model parameters, m

and

are the ﬁrst and second moment estimates, η is the

learning rate, and ε is a small constant to avoid divi-

sion by zero.

Gradient Clipping: To prevent the exploding

gradients problem, gradient clipping was applied,

where gradients are rescaled if their L2 norm exceeds

a threshold max norm:

clipped

= g ·min



max norm

∥g∥



(4)

Improving Disease Classiﬁcation Accuracy with Hybrid CNN-RNN Architectures for Lung Tumors

577

3.5 Training and Evaluation

The model utilized the Binary Cross-Entropy loss

function and was optimized using the Adam opti-

mizer. A dynamic learning rate adjustment strategy

was implemented through the ReduceLROnPlateau

scheduler, which halved the learning rate whenever

the validation loss showed no signiﬁcant improve-

ment.

To evaluate the performance of the model, met-

rics such as precision, recall, F1 score and the area

under the curve of the Receiver Operating Character-

istics (ROC) curve were used. Additionally, a 5-fold

cross-validation approach was employed to evaluate

the model’s ability to generalize effectively.

4 RESULTS

The hybrid CNN-RNN model was evaluated using a

validation dataset of 51,429 negative and 154 positive

samples. The model achieved the following perfor-

mance metrics: 94.8% recall, precision of 0.7602,

F1-score of 0.8434, and AUC of 0.901. These re-

sults, achieved with a validation accuracy of 96.1%,

demonstrate robust performance in detecting lung

nodules. Speciﬁcally, in the validation set, of the 154

cancerous nodules, our model has detected 148 cor-

rectly.

Comparison with Contemporary Models:

We compared the performance of our model with

several contemporary approaches from recent studies.

Table 1 highlights the performance metrics of these

models. The comparison underscores the superior re-

call and AUC of our hybrid CNN-RNN model rela-

tive to contemporary approaches. Recall, often con-

sidered critical in medical diagnostics, indicates the

model’s ability to correctly identify positive cases.

Our model’s recall of 94.8% surpasses all listed ap-

proaches, including the hybrid CNN-LSTM model

for Parkinson’s Disease (91%) and the Hybrid CNN-

RNN model for lung cancer survival (92%). This high

recall rate ensures minimal false negatives, a crucial

attribute when detecting potentially life-threatening

conditions like lung nodules.

Precision, which reﬂects the proportion of true

positive predictions among all positive predictions, is

another vital metric. Our model achieves a precision

of 0.7602, slightly lower than some other methods,

such as the Tomato Leaf Disease Detection model

(0.75) but consistent with other high-performing di-

agnostic systems. While precision could be further

optimized, the trade-off for higher recall is acceptable

in the context of critical health applications, where

missing positive cases is far more detrimental.

The F1-score provides a balanced view of the

trade-off between precision and recall. Our model’s

F1-score of 0.8434 is competitive, exceeding that

of the Hybrid CNN-LSTM for Parkinson’s Disease

(0.79) and the Respiratory Disease Prediction model

(0.79). This robust F1-score signiﬁes an effective bal-

ance in the model’s performance, making it suitable

for real-world clinical environments.

The AUC (Area Under the Curve) metric quanti-

ﬁes the model’s ability to distinguish between positive

and negative samples. Our model achieves an AUC of

0.901, which, while slightly lower than some other

methods such as the Hybrid CNN-RNN for Lung

Cancer Survival (0.97) and the Tomato Leaf Disease

Detection model (0.96), still demonstrates strong dis-

criminatory power. This solid AUC, combined with

the model’s high recall, underscores its reliability and

suitability for clinical diagnostics.

Table 1: Comparison of performance metrics with contem-

porary models

Model Precision Recall F1-Score AUC

Proposed Hybrid CNN-RNN Model 0.7602 0.9481 0.8434 0.901

Hybrid CNN-LSTM for Parkinson’s Disease (El-Sayed, 2024) 0.72 0.91 0.79 0.95

Tomato Leaf Disease Detection (Davida et al., 2022) 0.75 0.88 0.81 0.96

Hybrid CNN-RNN for Lung Cancer Survival (Lu et al., 2024) 0.68 0.92 0.78 0.97

Respiratory Disease Prediction (Li et al., 2024) 0.73 0.87 0.79 0.94

As shown in Table 1, our model demonstrates su-

perior recall and AUC, achieving a recall of 94.8%

and an AUC of 0.901. The balance between high re-

call and robust AUC highlights its potential for clini-

cal applications.

In medical diagnostics, recall is critical as it mea-

sures the model’s capacity to accurately identify pos-

itive cases. With a recall of 94.8%, our model out-

performs other approaches, such as the hybrid CNN-

LSTM model for Parkinson’s Disease (91%) and the

Hybrid CNN-RNN for lung cancer survival (92%).

This high recall minimizes false negatives, which is

essential in detecting serious conditions like lung nod-

ules.

Precision, an important indicator of how many

identiﬁed positive cases are genuinely positive, is an-

other key metric. Our model achieves a precision of

0.7602, comparable to similar systems, including the

Tomato Leaf Disease Detection model (0.75). While

there is room to enhance precision, the higher recall

justiﬁes this trade-off, particularly in health-related

applications where failing to identify true positives

could have severe consequences.

The F1-score, which balances precision and re-

call, further emphasizes our model’s effectiveness.

With a score of 0.8434, it surpasses other approaches

such as the Hybrid CNN-LSTM for Parkinson’s Dis-

INCOFT 2025 - International Conference on Futuristic Technology

578

ease (0.79) and the Respiratory Disease Prediction

model (0.79). This balanced performance demon-

strates the model’s suitability for deployment in clin-

ical settings, where both precision and recall are piv-

otal. Our model’s AUC of 0.901, while slightly be-

low the Hybrid CNN-RNN for Lung Cancer Survival

(0.97) and the Tomato Leaf Disease Detection model

(0.96), indicates a high level of reliability. The com-

bination of this solid AUC and exceptional recall un-

derscores the model’s strength and its potential for ap-

plication in real-world clinical diagnostics.

5 CONCLUSIONS

This proposed work presents a hybrid CNN-RNN

model for lung nodule detection, demonstrating

strong performance, particularly in recall and AUC,

which are crucial for identifying malignant nodules.

The model achieved an accuracy of 96.1% recall

of 94%, ensuring that most malignant cases are de-

tected, and an AUC of 0.901, indicating strong over-

all classiﬁcation performance. While the precision of

0.76 and F1-score of 0.84 are promising, there re-

mains room for improvement in reducing false pos-

itives, which can be achieved through further model

reﬁnement, threshold adjustments, and class balanc-

ing techniques.

The results underscore the importance of balanc-

ing sensitivity and speciﬁcity in medical imaging

models, especially when dealing with class imbal-

ances and small sample sizes, which are common in

lung cancer detection tasks. Future work could fo-

cus on integrating advanced data augmentation, semi-

supervised learning techniques, and more efﬁcient

preprocessing pipelines to further enhance precision

while maintaining high recall.

REFERENCES

Aslani, R. et al. (2022). Time-series deep learning model

for malignancy risk prediction in pulmonary nodules.

Computers in Biology and Medicine.

Chen, J. and Xie, R. (2024). Improved detection network

for lung nodule localization using deformable convo-

lution and self-paced learning. IEEE Transactions on

Medical Imaging.

Davida, H. E., Ramalakshmi, K., Venkatesan, R., and

Hemalatha, G. (2022). Tomato leaf disease detection

using hybrid cnn-rnn model. Journal of Applied Biol-

ogy, 65:134–145.

El-Sayed, R. S. (2024). A hybrid cnn-lstm deep learn-

ing model for classiﬁcation of the parkinson disease.

IEEE Access, 12:12345–12355.

et al., R. P. (2024). Ai-enhanced lung cancer detection

using the resnext50 architecture. In Proc. IEEE Int.

Conf. Imaging Systems (ICIS), pages 33–39. Avail-

able: https://ieeexplore.ieee.org/document/10627413.

Ewaidat, A. and El Brag, A. (2022). A convolutional neural

network-based approach using yolov5 for lung nodule

localization in ct scans. arXiv preprint.

Grand Challenge (2016). Luna16: Lung nodule analysis

2016.

Hesse, L. S. et al. (2020). Primary tumor origin classiﬁca-

tion of lung nodules in spectral ct using transfer learn-

ing. arXiv preprint.

Hosseini, M. et al. (2022). Deep learning applications for

lung cancer diagnosis: A systematic review. Journal

of Medical Imaging and Health Informatics.

Kumar, A. and Sharma, P. (2024). Deep learning based

lung cancer prediction using cnn. In Proc. IEEE

Int. Workshop on Machine Learning and Applications

(IWMLA), pages 14–20. Available: https://ieeexplore.

ieee.org/document/10627846.

Lee, J. and Gupta, H. (2023). Lung cancer diagnosis

and classiﬁcation using hybrid neural network tech-

niques. In Proc. IEEE Conf. Bioinformatics and Com-

putational Biology (BICB), pages 89–95. Available:

https://ieeexplore.ieee.org/document/10370424.

Li, L., Ayiguli, A., Luan, Q., Yang, B., and Subinuer, Y.

e. a. (2024). Prediction and diagnosis of respiratory

disease by combining cnn and bilstm methods. Jour-

nal of Healthcare Informatics, 12:112–120.

Liu, D. and Zhang, S. (2022). Lung cancer detection

using ct images and cnn algorithm. IEEE Access,

10:87766–87774. Available: https://ieeexplore.ieee.

org/document/9697158.

Liu, Y., Hou, Y.-J., Qin, C.-X., et al. (2023). A data aug-

mentation method and the embedding mechanism for

detection and classiﬁcation of pulmonary nodules on

small samples. arXiv preprint.

Lu, Y., Aslani, S., Zhao, A., Shahin, A., and Barber, D.

e. a. (2024). A hybrid cnn-rnn approach for survival

analysis in a lung cancer screening study. Journal of

Medical Imaging and Health Informatics, 14:35–44.

Marinakis, I., Karampidis, K., and Papadourakis, G. (2024).

Pulmonary nodule detection, segmentation and classi-

ﬁcation using deep learning: A comprehensive litera-

ture review. BioMedInformatics, 4(3):2043–2106.

Mehta, S. G. A. and Agarwal, V. (2024). Detection and ro-

bust classiﬁcation of lung cancer disease using hybrid

deep learning approach. In Proc. IEEE Int. Conf. Arti-

ﬁcial Neural Networks (ICANN), pages 27–35. Avail-

able: https://ieeexplore.ieee.org/document/10452545.

P. Mishra, T. S. and Kumar, R. (2024). Early lung cancer

detection using cnn. In Proc. IEEE Int. Symp. Compu-

tational Intelligence (ISCI), pages 98–104. Available:

https://ieeexplore.ieee.org/document/10627340.

Patel, A. G. R. and Sharma, K. (2024). A hybrid deep learn-

ing approach for early detection and classiﬁcation of

lung cancer using the pelican optimization algorithm.

In Proc. IEEE Int. Conf. Artiﬁcial Intelligence (ICAI),

pages 45–52. Available: https://ieeexplore.ieee.org/

document/10515355.

Improving Disease Classiﬁcation Accuracy with Hybrid CNN-RNN Architectures for Lung Tumors

579

Rao, M. and Patel, P. (2023). Lung cancer detection using

vgg16 and cnn. In Proc. IEEE Conf. Imaging and AI

(ICIA), pages 11–18.

Ronneberger, O., Fischer, P., and Brox, T. (2015). U-net:

Convolutional networks for biomedical image seg-

mentation.

Sharma, K. and Lee, J. (2023). Detection of lung can-

cer from pathological images using cnn model. In

Proc. IEEE Conf. Computational Pathology (CPath),

pages 91–98. Available: https://ieeexplore.ieee.org/

document/9574590.

Society, A. C. (2024). Lung cancer statistics. Ameri-

can Cancer Society. Online. Available: https://www.

cancer.org/research/cancer-facts-statistics.html.

Verma, R. A. S. and Singh, A. (2023). Ai-enhanced

lung cancer prediction: A hybrid model’s pre-

cision triumph. IEEE Trans. Med. Imaging,

43(6):1234–1240. Available: https://ieeexplore.ieee.

org/document/10643642.

Verma, S. and Kumar, A. (2023). Enhancing lung cancer

classiﬁcation: Leveraging existing cnns within a 1d

framework. In Proc. IEEE Conf. Data Science and

Engineering (ICDSE), pages 67–73. Available: https:

//ieeexplore.ieee.org/document/10695984.

Wang, C. et al. (2022). Towards reliable and explainable

ai model for solid pulmonary nodule diagnosis. arXiv

preprint.

Wang, N. and Li, Y. (2023). Revolutionizing lung cancer

prognosis through cnn-based predictive models. IEEE

J. Biomedical Engineering, 67(4):455–462.

INCOFT 2025 - International Conference on Futuristic Technology

580