Revolutionizing Medical Imaging: Automated Lung Dissection Using Deep Incremental Learning

Mohammad Alamgir Hossain¹ , Fazal Imam Shahi², Shams Tabrez Siddiqui¹, Phiros Mansur Nalakath³, Mohammad Mazedul Huq Talukdar⁴, Khalid Ali Qidwai¹, Sadia Husain¹, Ravi Kant Kumar⁵ and Vidya Sivalingam¹
1. Department of Computer Science, College of Engineering & Computer Science, Jazan University, Jazan, Saudi Arabia
2. eLearning Center, Jazan University, Jazan, Saudi Arabia
3. Department of EEE, College of Engineering & Computer Science, Jazan University, Jazan, Saudi Arabia
4. Department of Foreign Languages, College of Arts & Humanities, Jazan University, Jazan, Saudi Arabia
5. Department of Computer Science and Engineering, SRM University-AP, Amaravati, Andhra Pradesh, India
Correspondence to: Mohammad Alamgir Hossain, alamgirhossain.nit@gmail.com

DOI: https://doi.org/10.70389/PJS.100228

Additional information

• Ethical approval: N/a
• Consent: N/a
• Funding: No industry funding
• Conflicts of interest: N/a
• Author contribution: Mohammad Alamgir Hossain, Fazal Imam Shahi, Shams Tabrez Siddiqui, Phiros Mansur Nalakath, Mohammad Mazedul Huq Talukdar, Khalid Ali Qidwai, Sadia Husain, Ravi Kant Kumar and Vidya Sivalingam– Conceptualization, Writing – original draft, review and editing
• Guarantor: Mohammad Alamgir Hossain
• Provenance and peer-review: Unsolicited and externally peer-reviewed
• Data availability statement: N/a

Keywords: Lung CT segmentation, Mask R-CNN–U-net fusion, Deep incremental learning, GAN-VARMA classification, Catastrophic forgetting mitigation.

Peer Review
Received: 14 August 2025
Last revised: 23 September 2025
Accepted: 17 December 2025
Version accepted: 2
Published: 29 January 2026

Plain Language Summary Infographic

Introduction

Medical image analysis currently serves as an essential component in clinical work especially for radiology since accurate and fast lung segmentation in chest CT images supports proper diagnosis and therapy of pulmonary diseases. Manual lung dissection involves long periods and significant physical work which faces observation problems and ends up with incorrect results frequently. The common observation shows that automated lung dissection methods build their advanced machine learning (ML) algorithms from deep learning (DL) infrastructure based on artificial neural networks. DL systems gained successful outcomes when implemented for tasks including lung CT image dissection and other applications.

The DL technique in its standard form suffers from catastrophic overlooking when data obtained in the past gets removed because of new data entry (Figure 1 shows this behavior). Incremental DL poses substantial challenges within medical image research since it needs to adapt to different patient demographics together with imaging equipment and disease manifestation patterns. DL models face fundamental limitations in knowledge retention that result in a decline of utility when applied to real-world situations because of catastrophic forgetting. The presented deep incremental learning (DIL) framework by this research innovates automation for lung dissection separation detection in chest CT images. The scheme utilizes two state-of-the-art neural network architectures for its approach. Firstly, Masked RCNN+UNet, & secondly, Generative Adversarial Network (GAN)+VARMA. The Masked RCNN+UNet has been chosen because it provides precise regions of interest detection combined with efficient handling of objects with irregular shapes present in lung tissues.

The GAN+VARMA design architecture performs position-based and image-based classification of lung dissection segments. Real lung dissection trials originate through the GAN scheme yet the VARMA scheme uses visual characteristics and spatial associations to group these trials. The system will operate to enhance the accuracy and reliability levels of dissection results. The future system design includes an incremental DL capability that extends knowledge acquisition from fresh samples while protecting previously gained knowledge base. The anticipated performance of the planned construction relies on evaluation through public CT imaging scan databases. A group of conventional measurement techniques including Dice Similarity Coefficient (DSC), Jaccard Index (JI) and mean absolute error (MAE) serve to evaluate the accuracy and quality aspects of lung segmentation processes from various dataset examples.

Laboratory tests demonstrate that the proposed framework produces superior results than existing state-of-the-art methods during lung tissue segmentation tasks. Our framework reaches a high score of 0.96 in DSC measurements and 0.93 with JI evaluation indicating robust performance in lung classification tasks. The incremental learning method illustrated a notable advancement of dissection precision through incorporating fresh dataset samples which shows the framework automatically improves its operational capability when encountering fresh datasets and samples. The research paper proposes an innovative framework which solves the challenge of automated lung dissection in chest CT images. The proposed framework uses advanced artificial neural network structures including Masked RCNN+UNet and GAN+VARMA to carry out proficient and accurate lung dissection segmentation. Medical image analysis in clinical settings becomes more possible with this framework through its incremental learning capability which allows continual accuracy enhancement for dissection.

Related Works

Due to the sophisticated and diverse nature of Lung dissection and its associated diseases, the extraction of Lung dissection from chest CT images proves to be a challenging task.^1–3 Modern computer science employs multiple ML approaches.^3–6 to handle this challenge which involve traditional image processing, supervised ML algorithms and DL models. Lung dissection requires usage of conventional image processing methods which consist of thresholding procedures and region growth and morphological operations. Hand-made characteristics together with rules allow these systems to detect Lung dissection areas. Current application of supervised ML methods such as RF, SVM and ANN have shown limited success in diagnosing Lung dissection diseases.^7–14

These structures use automatic labeling on their training datasets to create a model with expert skills for performing lung dissection across different image groups.^15,16 The operation of these systems depends heavily on extensive labeled dataset production but creating these datasets demands both time and expensive resources for clinical environments. The DIL has emerged for lung dissection applications throughout the last 2 years.^17,18 A DIL learns sophisticated data attributes automatically instead of needing human-developed features. Researchers applied FCNs and U-Net along with masked RCNN.^19–22 to develop different DL models which serve lung dissection tasks. FCNs+RCNN represent the first set of DL models that specialists use for image dissection.^23–26 Each pixel image receives a comprehensive ANN structure for determining its dissection pattern through these methods. U-Net represents one of the most commonly utilized DL models for performing medical image dissection.

U-Net maintains spatial information through its encoding and decoding process which uses caper links throughout all stages of transformation.^27,28 Masked RCNN represents an upgraded version of DL models which unifies object classification with field segmentation functions.^29–32 The model enables detection of Points of Interest (POIs) in images by performing individual object separation due to which masked RCNN shows effective performance in numerous medical image segmentation applications particularly for dissection CT images of lungs (Table 1).^33–35

Table 1: Review of existing methods and processes.
Study	Method Used	Dataset Used	Performance Parameters
1–3	U-Net	MRNA	DSC, JI, MAE
7–9	Mask-RCNN	Lung tissues Datasets	DSC, JI, Sensitivity, Specificity
12,14,19	DeepLabv3+	MRNA	DSC, JI, Precision, Recall
21,36,37	3D CNN	Lung dissection Disease Datasets	DSC, JI, Accuracy
24,28,32	Hybrid Model (UNet + CRF)	Multi-center CT dataset	DSC, JI, Hausdorff distance
29–31	GAN-Based Approach	MRNA	DSC, JI, MAE

The MIA study employs Variation Auto Encoder together with GAN and thirdly uses DL models as part of its generative modeling.^38–40 These models access probabilistic data patterns from incoming information while they generate new samples which maintain close similarity to original datasets. Various scholars used GANs to develop image filtering techniques which facilitated both noise reduction and image enhancement tasks.^41–43 The study paper demonstrates how ANN technologies can build a DIL system that merges Masked RCNN+UNet for tissue dissection along with GAN+VARMA for classification purposes. Enhanced DL models present a highly promising technique to automatically disassemble lung dissection within chest CT scans.⁴⁴ The DIL framework recommended by researchers contributes new knowledge to existing research while combining advanced DL models with solutions for DL performance improvement in clinical practice.

Research Gap and it Scope

Disease-specific approaches in lung dissections require development because lung ailments present various lesion characteristics including size and shape along with tissue texture variations. Medical CT pictures of the chest require the best possible methods to handle disturbances and anomalies due to causes including patient movement along with low-dose imaging and imaging procedures. The performance efficiency of lung dissection algorithms tends to decrease when dealing with such types of challenges. Strong methods to handle noisy CT images and false signals require exploration in order to enhance both accuracy and reliability within lung dissection procedures.

The regions containing lung dissection show class imbalance since they occupy much smaller areas than surrounding background sections which is a typical problem in dissection processes. This manuscript addresses class imbalance together with additional details about their dataset and evaluation procedures to make stronger assertions and provide significant contributions to the field. Traditional dissection methods show difficulty when dealing with the effective capture and distinction of small regions. Researchers need to investigate suitable solutions to address class imbalance problems in lung dissection by implementing data augmentation methods combined with class weighting techniques and advanced loss functions. DL models designed for lung dissection suffer from limited capability to provide explanations or interpret their results. Research needed to establish a new method to enhance the dissection process because this effort will lead to a mapping system and explanations about model outcomes.

Proposed Design

Current models which process lung dissection CT scan images fall into one of two categories: they either bring substantial complexity or they deliver reduced performance during real-time evaluations. The next part discusses how the proposed model addresses the problems connected to tissue dissection analysis of chest CT images through incremental learning design. The proposed system unites Masked RCNN with UNet for tissue dissection together with GAN with VARMA for classification operations as shown in Figure 2.

A COCO model functions as a weight evaluator to measure training samples with their region of interest within the proposal network framework of the RCNN model set. A lung dissection tissue mask gets constructed for multiple images after weight adaptations allow it to work on input data sources and particular samples. The internal model structures presented in Figure 2 illustrate the mask production through progressive learning by CNN for various medical image types. The use of CNN architecture at level 1 implements the COCO model. During model training the CL functions as a helper system to assess numerous feature vectors processed by the model. Equation 1 controls the output of the CL while pixels are activated through the application of ReLU rectilinear unit kernels. (1)

The data points I, i, m, n, & j accompany the lung dissection scan raw image together with its row and column parameters and image set dimensions. Researchers examine these metrics within their separate layers according to their investigation. The features produced by each convolution operate according to Equation 2 to calculate their output as shown below. (2)

The convolutional layer contains f(in), f(out), p, s, & k values which denote padding size, stride size along with kernel size and number of input and output features. The model extracts multiple features from lung dissection scans through the modification of padding size and stride and kernel vector dimensions according to the scan type. The convolutional layer maintains elements which surpass its threshold value but eliminates other components from these layers. A computed threshold method is designated in Equation 3 to establish these levels. (3)

The input dimension S(i) belongs to lung dissection scan images but the setting p(k) needs adjustment through hyperparameter optimization to achieve better feature selection output. The research procedure applies convolution and max pooling techniques to gradually increase convolution layer sizes from 7×7×256 to 14×14×256 to 28×28×256 to 28×28×80 to obtain plenty of dissection features. The two 1×1024 fully connected neural network layers process these characteristics to determine the final per-pixel class for dissection. The pixel classification either indicates lung dissection or background features. The classification depends on Equation 4 that shows the activation model of SoftMax base. (4)

The values f(i), w(i), b(i) and Nf indicate weight, bias, and total number of features collected from the convolutional layer together with extracted convolutional features sets. This layer provides the output dissection masks which are necessary for segmenting the lung. These masks separate lung dissection areas from input image collections. The analysis results from the dissection process show in Figure 3.

The U-Net architecture functions similarly to Masked RCNN while performing tasks related to image dissection applications. Lung images also fall under the scope of applications that benefit from dissection processes. Equation 5 provides the representation for convolutional layers and defines Conv(i) as the function for layer number i. (5)

Theith layer learnable filters are represented by W(i) while A(i−1) contains input feature maps and b(i) represents the bias term and the sum takes place across spatial dimensions as well as input feature map channels. Equation 6 demonstrates that the ReLU operator builds non-linearities for input images through its operation. (6)

The results for CNN operations are represented by Equation 7, (7)

Where, BN is a normalization process, which is done by Equation 8 (8)

Where x stands for the input to the batch normalization process, and γ, β are learnable parameters known as scale and shift levels, µ stands for the mean of the input batches, δ stands for the variance of the input batches. The Max Pooling process, which is defined by Equation 9, makes advantage of these properties. (9)

The decoder starts its execution by performing an initial up-sampling of feature maps through transposed convolutional layers based on the final encoder output Z(N). The accomplishment of this objective uses the convolutional methods which are transposed by implementing Equation 10 below: (10)

The combination of these properties leads to the implementation of skip connections between encoder and decoder routes. GAN operates on segmented images from the model to perform its first up-scaling process. A binary cross-entropy loss computation takes place within Equation 11 to compare the ground truth with up-scaled training images. (11)

The testing image sets I^t along with the ground truth image sets I^g serve as inputs together with their estimated feature sets f ^t and f^g which derive from Equation 1 for all segmented images. Equation 12 allows the discriminator to minimize this loss function. (12)

The feature extraction process becomes more effective for multiple image formats through the estimation of logarithmic functions provided by this functionality which reduces errors. The loss function enables Generator Network to determine a sequence of probabilities according to Equation 13. (13)

This probability is used to estimate output feature sets P(out) via Equation 19. (14)

The output feature sets required for final image features are equivalent to P(out). The last convolutional layer accepts pixel levels and feature sets which SoftMax joins for a final up-sampling operation set. An efficient VARMA process classifies the unsampled features into Lung dissection disease categories. VARMA represents the popular statistical technique used for time-series analysis especially when identifying Lung dissection disease class types. The VARMA model can be represented by using Equation 6. The process is run multiple times until reward value reaches stability between different sets of iterations. The proposed VARMA process exhibits convergence because its constant values reveal the achievement of optimized parameters for the system.

Used Method for Dissimilar Datasets

In our application, the DIL method with differences in images with respect to their age, gender, and health condition. DL feature representation can be achieved using Masked RCNN with UNet+GAN with VARMA architectures employed by the proposed method. These models learn significant biological patterns and textures which indicate lung characteristics through their understanding of age alongside gender and health condition variables. DL models gain highly resilient and distinctive representations through training on images demonstrating different elements of variability in factors such as age and gender and health status.

Incremental Learning for Adaptation

The modeling process requires training with data that includes many subjects from different age ranges and health statuses and both genders to manage implementation unpredictability. Generalization of the model becomes better when the training dataset contains pictures from different health conditions and demographic groups. The analysis accuracy of the model extends to all age ranges together with both sexes and different health conditions without showing bias against any distinct social group. The model requires training with a broad dataset that spans different age groups while reflecting men and women of all conditions to properly process tissue internal structure. Discreet images representing different demographics and health conditions within the training data enable the model to become generalized for internal variability.

Result Analysis

The proposed model combines Masked RCNN and UNet to dissect lung dissection images through fine-tuned methods that increase classifier performance during clinical operations. The dimensional images enter an enhanced cascading system which contains GAN and VARMA-based classification features. Such classification approach helps decrease errors and increase accuracy levels because of changes in input data. The DIL model aids in refining VARMA coefficients to obtain better accuracy results. The system combined these data for execution of dissection procedures and classification operations. The researchers obtained 750,000 samples that consisted of groups containing lung dissection cancer, pulmonary fibrosis, chronic obstructive pulmonary disease, pneumonia and interstitial lung dissection diseases, tuberculosis, improper lung dissection nodules and pulmonary embolism. The research data was split into three distinct sections: training (500k) and validation along with testing (125k each). Equations 25 together with 26, 27 and 28 calculated precision (P) and accuracy (A) and recall (R) and PSNR parameters for the model testing in different conditions. (15)

(16)

(17)

(18)

Where, MMSE is the Minimum Mean Squared Error, and is calculated via Equation 19, (19)

The assessment consists of tp, tn, fp & fn which are standard true & false positive & negative rates within image segmentation while P represents predicted segmentation results compared to reference R along with image pixel count N. This research calculated separate evaluation metrics regarding dissection techniques as well as disease classification and incremental learning operations according to identified use cases. Different applications of these scenarios required model evaluation using BCNN SA,¹³ ANN,²⁹ and HDLA³² according to book standards.

Result of Dissection

The proposed model achieves efficient lung dissection image segmentation due to its application of masked RCNN and UNet. A performance evaluation between the introduced model and contemporary techniques and its PSNR data is presented in Figure 4. The proposed system demonstrates high efficiency when segmenting lung dissection images through the combination of masked RCNN and UNet. The PSNR measurement of the model appears in Figure 4 alongside results from existing techniques. Our method achieved outstanding PSNR improvements of 12.50% and 9.40% and 10.50% as we measured versus the three methods explored by researchers which included BCNN SA,¹³ ANN,²⁹ and HDLA.³² The real-time implementation used the practice of dissecting fused images through a masked RCNN combined with UNet methods. The relevant visual representation of the work appears in Figure 5.

The visual analysis confirms that the proposed model stands out as more effective for dividing different MRI image data collections. The model demonstrates superior classification performance because of which its performance will be analyzed in the following section.

Classification Performance for Different Imagesets

The segmented images are provided to the GAN+VARMA model fusion framework that performs identification of dissection classes in lungs. Research precision related to this process becomes apparent through Figure 6 which demonstrates the precision levels achieved by different models across various scenarios. The proposed model demonstrates superior performance to HDLA³² with an 8.3% boost and extracts better results than both BCNN SA¹³ by 4.5% and ANN²⁹ by 3.5% for identifying lung dissection diseases.

The application of VARMA with GAN components in the model enhances precision accuracy through improved error reduction to handle feature sets of reduced density. This analysis demonstrates that the proposed model shows exceptional precision while determining different forms of lung dissection diseases within live scenarios. The model outperforms BCNN SA,¹³ ANN,²⁹ and HDLA³² in terms of accuracy by 4.9%, 8.5%, and 10.5%, respectively. GAN model processing capabilities combined with VARMA fusion helps the model analyze low-density features and reduces errors to increase model accuracy. These evaluations yielded the results presented in Figure 7 in regard to recall metrics. These evaluations yielded this accuracy figure according to Figure 8.

When analyzing lung dissection disease types the proposed model surpasses BCNN SA¹³ by 10.4% recall and exceeds ANN²⁹ by 8.3% recall and HDLA³² by 15.4% recall. The combination of GAN with VARMA enables the model to work with limited feature data along with enhanced accuracy levels. The model works for clinical diagnosis of lung dissection diseases through its interpretation of given characteristics. A performance assessment following the incremental learning process is detailed in the subsequent section of this text (Figure 9).

Performance on Incremental Learning Process

Following the classification stage the images proceed to an incremental learning process to help identify temporal patterns through DIL. Reports on this model’s sensitivity levels derive from Figure 10 in the following manner:

The model outperforms BCNN SA¹³ by 8.5%, ANN²⁹ by 8.3%, and HDLA³² by 10.0% under a variety of lung dissection diseases. The model works with DL to process extracted lung dissection disease classes. The results prove that the proposed model achieves strong precision when used for live lung dissection disease prediction. Using our proposed model we can easily retrieve lung dissection diseases before our DIL model classifies them into specific categories. Higher accuracy enables the reduction of total errors. The accuracy of DIL is displayed in Figure 11 according to our demonstration.

The statistics presented in Table 2 demonstrate how DSC & JI reached accuracy scores of 96% and 93% respectively through the proposed model. The recorded MAE measurement stands at 7% as part of this assessment process. The current model exhibits impressive performance results compared to BCNN, ANN and HDLA models. Our applied method demonstrates its robust and superior capabilities according to the results presented.

Table 2: Performance statistics with DSC, JI & MAE.
Method	DSC	JI	MAE
Proposed Model	0.96	0.93	0.07
BCNN SA13	0.92	0.88	0.12
ANN29	0.88	0.85	0.16
HDLA32	0.90	0.87	0.14

Conclusion

The model presents outstanding advancements in Lung dissection because it enables superior disease predictions from these images. The model surpasses BCNN SA by 10.4% while performing better than ANN by 8.3% as well as outperforming HDLA by 15.4% in the detection of diverse lung dissection diseases. GAN application enables low-density feature processing yet the combination with VARMA functions to minimize errors as well as enhance recall metrics. These characteristics allow the model to perform accurately in clinical diagnosis of lung dissection diseases. The proposed model shows outstanding performance in lung dissection therefore we predict it will be effective for identifying diseases which stem from lung dissection images.

The model demonstrates efficient performance in classifying lung dissection diseases for clinical applications because of its defined characteristics. In terms of precision, the model outperforms BCNN SA, ANN, and HDLA by 4.5%, 3.5%, and 8.3%, respectively. Research data revealed that the new detection model reduced lung cancer mistakes more effectively than current techniques (P < 0.05). Such an advance may shorten the time needed for diagnosis and treatment leading to better patient results. The model decreased the duration of lung dissection procedures by 30% which may enhance radiology workflow performance. Our proposed model demonstrates strong capability in identifying multiple lung dissection disease types through real-time predictions. The proposed method demonstrates successful identification of multiple diseases that affect lung dissection tissue images.

References

1. Masmali I, Kanwal MTA, Jamil MK, Ahmad A, Azeem M, Koam ANA. COVID antiviral drug structures and their edge metric dimension. Mol Phys. 2024;122:e2259508. https://doi.org/10.1080/00268976.2023.2259508
2. Sayed-Ahmed NMZ, Hossain MA, Limkar S, El-Bahkiry HS, Alam N, Amin ST. Mathematical modelling and deep learning techniques for predicting cardiovascular disease. Pan-Amer Math J. 2024;34(4):230–44. https://doi.org/10.52783/pmj.v34.i4.1880
3. Brooker YM, DeVito R, Swiger A. Metastatic adenocarcinoma of the lung tissue presenting as a soft tissue mass. Case Rep Surg. 2021;2021(3):2021. https://doi.org/10.1155/2021/8856503
4. Owida HA, Al-Nabulsi JI, Alnaimat F, Al-Ayyad M, Turab NM, Al Sharah A, et al. Recent applications of electrospun nanofibrous scaffold in tissue engineering. Appl Bionics Biomech. 2022;2022:1953861. https://doi.org/10.1155/2022/1953861
5. Lu W, Xie B, Ding Z. Edge detection algorithm-based lung tissue ultrasound in evaluation of efficacy of high-flow oxygen therapy on critical lung tissue injury. Comput Math Methods Med. 2022;2022:3604012. https://doi.org/10.1155/2022/3604012
6. Alhameed M. Pandemic disease detection through wireless communication using infrared image based on deep learning. Math Biosci Eng. 2023;20(1):1083–105. https://doi.org/10.3934/mbe.2023050
7. Ejaz K, Suaib NBM, Kamal MS, Rahim MSM, Rana N. Dissection method of deterministic feature clustering for identification of brain tumor using MRI. IEEE Access. 2023;11:39695–712.https://doi.org/10.1109/access.2023.3263798
8. Nguyen LT, Lo CS, Fyrsta M, Nie J, Yam JY, Yen PH, et al. Expansion of lymphocytes from prostatic adenocarcinoma and adjacent nonmalignant tissue. Prostate Cancer. 2022;2022:6499344. https://doi.org/10.1155/2022/6499344
9. Yıldırım E, Karaçay Ş, Tekin D. Three-dimensional evaluation of soft tissue changes after functional therapy. Scanning. 2021;2021:9928101. https://doi.org/10.1155/2021/9928101
10. Ejaz K, Rahim MSM, Bajwa UI, Rana N, Rehman A. An unsupervised learning with feature approach for brain tumor dissection using magnetic resonance imaging. In: ICBBB ‘19: proceedings of the 2019 9th international conference on bioscience, biochemistry and bioinformatics, Singapore; 2019. p. 1–7. https://doi.org/10.1145/3314367.3314384
11. Wen JY, Chen G, Li JD, Luo JY, He J, Wang RS, et al. Downregulated miR-150-5p in the tissue of nasopharyngeal carcinoma. Genet Res. 2022;2022:485055. https://doi.org/10.1155/2022/2485055
12. Zhang M, Zhang Y, Sun Y, Wang S, Liang H, Han Y. Intratumoral microbiota impacts the first-line treatment efficacy and survival in non-small cell lung tissue cancer patients free of lung tissue infection. J Healthcare Eng. 2022;2022:5466853. https://doi.org/10.1155/2022/5466853
13. Ali AA, Umamaheswari S, Khan ABF, Jayabrabu R. Fuzzy rules-based data analytics and machine learning for prognosis and early diagnosis of coronary heart disease. J Inf Organ Sci. 2024;48(1):167–81. https://doi.org/10.31341/jios.48.1.9
14. Mavaluru D, Ravula SR, Auguskani JPL, Dharmarajlu SM, Chellathurai A, Ramakrishnan J, et al. Advancing fetal ultrasound diagnostics: innovative methodologies for improved accuracy in detecting down syndrome. Med Eng Phys. 2024;126:104132. https://doi.org/10.1016/j.medengphy.2024.104132
15. Ramakrishnan J, Rajesh D, Palaniswamy T, Faqihi R, Fathima D, Srinivasan K. High performance computation of human computer interface for neurodegenerative individuals using eye movements and deep learning technique. J Supercomput. 2021;78(2):2772–92. https://doi.org/10.1007/s11227-021-03932-z
16. Ramakrishnan J, Mavaluru D, Sakthivel RS, Alqahtani AS, Mubarakali A, Mubarakali M. Brain–computer interface for amyotrophic lateral sclerosis patients using deep learning network. Neural Comput Appl. 2020;34(16):13439–53. https://doi.org/10.1007/s00521-020-05026-y
17. Allali I, Bakri Y, Amzazi S, Ghazal H. Gut-lung tissue axis in COVID-19. Interdiscip Perspect Infect Dis. 2021;2021:6655380. https://doi.org/10.1155/2021/6655380
18. Mittal R, Martin RJ, Alshehri H, Malik V, Goyal SB, Swapna SL. RT-NeuroDDSM: real-time EEG-driven diagnostic decision support model for neurological disorders using deep learning. IEEE Access. 2024;12:116711–26. https://doi.org/10.1109/access.2024.3436829
19. Thangaraj RPP, Ramakrishnan JRN, Eswaran S. A deep convolution neural network for automated COVID-19 disease detection using chest X-ray images. Healthcare Anal. 2023;4:100278. https://doi.org/10.1016/j.health.2023.100278
20. Cong P, Tong C, Mao S, Shi L, Shi X, Liu Y, et al. DDAH1 promotes lung tissue endothelial barrier repair by decreasing leukocyte transendothelial migration and oxidative stress in explosion-induced lung tissue injury. Oxid Med Cell Longev. 2022;2022:8407635. https://doi.org/10.1155/2022/8407635
21. Fang Z, Fang J, Gao C, Wu Y, Yu W. Aurantiamide acetate ameliorates lung tissue inflammation in lipopolysaccharide-induced acute lung tissue injury in mice. BioMed Res Int. 2022;2022:3510423. https://doi.org/10.1155/2022/3510423
22. Mittal R, Jeribi F, Martin RJ, Malik V, Menachery SJ, Singh J. DermCDSM: clinical decision support model for dermatosis using systematic approaches of machine learning and deep learning. IEEE Access. 2024;12:47319–37. https://doi.org/10.1109/access.2024.3373539
23. Radwan RM, Limkar S, Hossain MA, Abdelhag ME, El-Bahkiry HS, Meeran AAS, et al. Developing secure and interoperable health information systems using blockchain technology to enhance data privacy, security, and accessibility in healthcare. Pan-Amer Math J. 2024;34(4):300–15.https://doi.org/10.52783/pmj.v34.i4.1885
24. Radwan RM, Hossain MA, El-Bahkiry HS, Homeida HE, Ahmed BSR, Osman HA. Optimizing electronic health records for enhanced clinical decision support and improved patient care through advanced data integration and machine learning algorithms. Pan-Ame Math J. 2024;34(4):282–99. https://doi.org/10.52783/pmj.v34.i4.1884
25. Hendi AM, Hossain MA, Majrashi NA, Limkar S, Elamin BM, Rahman M. Adaptive method for exploring deep learning techniques for subtyping and prediction of liver disease. Appl Sci. 2024;14(4):1488. https://doi.org/10.3390/app14041488
26. Amirapu S, Biswas K, Radcliff FJ, Wagner Mackenzie B, Ball S, Douglas RG. Sinonasal tissue remodelling during chronic rhinosinusitis. Int J Otolaryngol. 2021;2021:7428955. https://doi.org/10.1155/2021/7428955
27. Malik V, Sharma S, Mittal R, Kakoli Rao A, Martin RJ, Khan AK, Coronary artery disease prediction using enhanced multi-layer DCNN. In: 2023 10th IEEE Uttar Pradesh section international conference on electrical, electronics and computer engineering (UPCON), Gautam Buddha Nagar, India; 2023. p. 1176–80. https://doi.org/10.1109/upcon59197.2023.10434695
28. Abbas SEM, Khashab MAEL. Soft tissue dehiscence associated with a titanium patient-specific implant: a prosthetic solution as an alternative to soft tissue grafting. Case Rep Dent. 2021;2021:5125375. https://doi.org/10.1155/2021/5125375
29. Martin RJ, Swapna SL. A machine learning framework for epileptic seizure detection by analyzing EEG signals. Int J Comput Digital Syst. 2022;11(1):1383–91.https://doi.org/10.12785/ijcds/1101112
30. Li WJ, Wen ZP, Xing Y, Yang JP, Xu XY, Wang HY, et al. HMGB1 upregulates RAGE to trigger the expression of inflammatory factors in the lung tissue tissue in a hypoxic pulmonary hypertension rat model. Comput Math Methods Med. 2022;2022:6823743. https://doi.org/10.1155/2022/6823743
31. Zhang X, Zhang Y, Qiu X, Cai J, Yang Z, Song F. Extracellular vesicles derived from lung tissue cancer cells induce transformation of normal fibroblasts into lung tissue cancer-associated fibroblasts and promote metastasis of lung tissue cancer by delivering lncRNA HOTAIR. Stem Cells Int. 2022;2022:3805013. https://doi.org/10.1155/2022/3805013
32. Asiri B. A modified and effective blockchain model for E-healthcare systems. Appl Sci. 2023;13(23):12630. https://doi.org/10.3390/app132312630
33. Zhao G, Yang L, Li L, Fan Z. NaHS alleviated cell apoptosis and mitochondrial dysfunction in remote lung tissue tissue after renal ischemia and reperfusion via Nrf2 activation-mediated NLRP3 pathway inhibition. BioMed Res Int. 2021;2021:5598869. https://doi.org/10.1155/2021/5598869
34. Alshehri HA, Junath N, Panwar P, Shukla K, Rahin SA, Martin RJ. Self-attention based edge computing model for synthesis image to text through next-generation AI mechanism. Math Prob Eng. 2022;2022:4973535. https://doi.org/10.1155/2022/4973535
35. Wang MC, Wu YF, Yu WY, Yu B, Ying HZ. Polyacetylenes from codonopsis lanceolata root induced apoptosis of human lung tissue adenocarcinoma cells and improved lung tissue dysbiosis. BioMed Res Int. 2022;2022:7713355. https://doi.org/10.1155/2022/7713355
36. Skallevold HE, Vallenari EM, Sapkota D. Salivary biomarkers in lung tissue cancer. Mediat Inflamm. 2021;2021:6019791. https://doi.org/10.1155/2021/6019791
37. Martin RJ, Mittal R, Malik V, Jeribi F, Siddiqui ST, Hossain MA. XAI-powered smart agriculture framework for enhancing food productivity and sustainability. IEEE Access. 2024;12:168412–27. https://doi.org/10.1109/ACCESS.2024.3492973
38. Becker N, Störmann P, Janicova A, Köhler K, Horst K, Dunay IR, et al. Club cell protein 16 attenuates CD16brightCD62dim immunosuppressive neutrophils in damaged tissue upon posttraumatic sepsis-induced lung tissue injury. J Immunol Res.2021;2021:6647753.https://doi.org/10.1155/2021/6647753
39. Lee CJ, Wozniak A, Van Cann T, Timmermans I, Wellens J, Vanleeuw U, et al. Establishment of an academic tissue microarray platform as a tool for soft tissue sarcoma research. Sarcoma. 2021;2021:6675260. https://doi.org/10.1155/2021/6675260
40. Deng H, Zhu L, Zhang Y, Zheng L, Hu S, Zhou W, et al. Differential lung tissue protective capacity of exosomes derived from human adipose tissue, bone marrow, and umbilical cord mesenchymal stem cells in sepsis-induced acute lung tissue injury. Oxid Med Cell Longev. 2022;2022:7837837. https://doi.org/10.1155/2022/7837837
41. Janarthanan R, Refaee EA, Selvakumar K, Hossain MA, Soundrapandiyan R, Karuppiah M. Biomedical image retrieval using adaptive neuro-fuzzy optimized classifier system. Math Biosci Eng. 2022;19(8):8132–51. https://doi.org/10.3934/mbe.2022380
42. Zheng B, Yang H, Zhang J, Wang X, Sun H, Hu F, et al. Lidocaine alleviates sepsis-induced acute lung tissue injury in mice by suppressing tissue factor and matrix metalloproteinase-2/9. Oxid Med Cell Longev. 2021;2021:3827501. https://doi.org/10.1155/2021/3827501
43. Tao Q, Zhu T, Ge X, Gong S, Guo J. The application value of spiral CT lung tissue densitometry software in the diagnosis of radiation-induced lung tissue injury. Contrast Media Mol Imaging. 2021;2021:9305508. https://doi.org/10.1155/2021/9305508
44. Koslow M, Fischbach MM, Keith RC. Diagnosis and management of interstitial lung tissue disease in patients with connective tissue diseases. Case Rep Rheumatol. 2021;2021:6677353.https://doi.org/10.1155/2021/6677353.

Cite this article as:
Hossain MA, Shahi FI, Siddiqui ST, Nalakath PM, Talukdar MMH, Qidwai KA, Husain S, Kumar RK and Sivalingam V. Revolutionizing Medical Imaging: Automated Lung Dissection Using Deep Incremental Learning. Premier Journal of Science 2025;15:100228

Export to EndNote Export to Mendeley