A Comparative Review of Machine Learning and Deep Learning Models for Lung Disease Classification Across Medical Imaging Modalities

Abstract

The increasing integration of artificial intelligence into medical imaging has significantly advanced lung disease detection using chest X-rays and related modalities. This survey conducts a comprehensive comparative analysis of 12 researchworkspublishedbetween 2021 and 2025, with emphasis on five key studies employing machine learning, deep learning, and hybrid algorithms for image-based lung disease classification. Design/methodology/approach–Thereviewexaminesmethodologiesinvolving segmentation-classification pipelines, attention-enhanced CNNs, hybrid CNN–ELM architectures, multi-branch deep networks, and fusion models combining handcrafted and deep features. Datasets used across these studies include NIH ChestX-ray14, JSRT, MC, LC25000, SIIM-ACR, LIDC-IDRI,andCOVID-19imagerepositories.Eachalgorithmisanalyzedinterms of preprocessing strategies, model design, evaluation metrics, computational efficiency, and applicability to different imaging types such as chest X-rays, CT scans, and histopathologyslides. Findings – The analysis shows thatnosinglearchitectureperformsbestacrossalltasks;instead, each model demonstrates unique advantages depending on data modality and clinical objective. Attention-based CNNs excel in feature refinement, hybrid CNN-ELM models offer lightweight computation, and dual-branch frameworks improve multi-label chest disease detection.. Originality/value – By consolidating state-of-the-art image-based lung disease classification research, this survey highlights methodological strengths, limitations, and emerging trends. Although key research gaps are identified for future inclusion, the present review provides a structured foundation for developing more scalable and reliable diagnostic systems.

Introduction

Lung diseases—including COVID-19, pneumonia, lung cancer, fibrosis, emphysema, and related thoracic conditions—remain a major global health burden, with chest X-ray (CXR) and CT scans as primary diagnostic tools. Manual interpretation of these images is challenging and prone to variability, especially in high-volume clinical settings. Recent advances in machine learning (ML) and deep learning (DL) have enabled automated, data-driven analysis of medical images, significantly improving diagnostic accuracy, sensitivity, and speed through models such as CNNs, attention mechanisms, hybrid frameworks, detection networks, and transformer-based architectures.

Despite rapid progress, existing reviews are limited by a narrow focus on lung cancer, single imaging modalities, or lack of systematic performance comparison across diverse algorithms. To address these gaps, this review presents a comprehensive comparative analysis of 12 ML/DL studies published between 2021 and 2025, covering multiple lung diseases and imaging modalities including CXR, CT, histopathology, and PET/CT. It provides a unified summary of model architectures, preprocessing and segmentation strategies, performance metrics, and limitations, along with a structured comparison across algorithm families.

Using a PRISMA-inspired methodology, studies were selected from IEEE Xplore, Google Scholar, and SpringerLink based on defined inclusion and exclusion criteria. The review highlights how imaging modality strongly influences model design: CXR favors lightweight and attention-based CNNs, CT benefits from segmentation and detection pipelines, histopathology requires deep and patch-based architectures, and PET/CT demands multimodal fusion strategies. Overall, the review clarifies the strengths and limitations of current ML/DL approaches and offers guidance for selecting suitable algorithms for specific lung diseases and imaging contexts.

Conclusion

This review presented a comprehensive comparative analysis of twelve recent machinelearning and deep learning–based studies (2021–2025) addressing lung disease prediction and classification using image data. By organizing existing work through a clear algorithmic taxonomy and evaluating models across CXR, CT, histopathology, and PET/CT modalities, the study highlights that no singlemodelisuniversallyoptimalforalldiseasesandimagingsettings. Instead, performance is strongly influenced by the choice ofarchitecture,imagingmodality,and diagnostic task. Attention-enhanced CNNs and hybrid architectures demonstrated consistently strong performance by refining discriminative features, while detection-oriented models suchas YOLO and RetinaNet proved particularly effective for lesion localization tasks. CT- and histopathology-based systems achieved the highest reported accuracies overall, benefiting from richerspatialandmorphologicalinformation,whereasPET/CTfusionmodelsshowedpromisein capturing complementary metabolic and structural cues. Despite these advances, the analysis also reveals critical challenges that must be addressedbefore widespread clinical adoption is feasible. Many studies rely on limited, single-center datasets, leading to generalization and dataset bias issues that restrict real-world applicability. Disease-level insights indicate that while current models perform well for common conditions such as pneumonia and COVID-19, detecting small nodules and rare cancer subtypes remains challenging. Furthermore, clinical deployment requires greater robustness, interpretability, and validation, as most modelslackclinician-in-the-loopevaluationandtransparentdecision-making mechanisms. Moving forward, the development of standardized datasets, explainable and trustworthy AI systems, and computationally efficient models suitable for real hospital environments will be essential. By consolidating state-of-the-art methodologies, performance trends, and open challenges, this review provides a structured foundation for future research aimed at building scalable, reliable, and clinically meaningful AI-driven lungdiseasediagnostic systems.

References

[1] Li,Kai,etal.\"MI2A:AMultimodalInformationInteractionArchitectureforAutomatedDiagnosis of Lung Nodules Using PET/CT Imaging.\" IEEE Sensors Journal (2025). [2] Ozdemir,Burhanettin,EmrahAslan,andIshakPacal.\"Attentionenhancedinceptionnextbased hybrid deep learning model for lung cancer detection.\" IEEE Access (2025). [3] Chaddad,Ahmad,JihaoPeng,andYihangWu.\"Classificationbaseddeeplearning models forlungcanceranddiseaseusing medicalimages. \"IEEE TransactionsonRadiationandPlasmaMedical Sciences (2025). [4] Agarwal, Saurabh, K. V. Arya, and Yogesh Kumar Meena. \"Cnn-o-elmnet: Optimizedlightweight andgeneralizedmodel forlungdiseaseclassificationandseverityassessment.\"IEEETransactions on Medical Imaging 43.12 (2024): 4200-4210. [5] Aharonu, Mattakoyya, and Lokeshkumar Ramasamy. \"A multi-model deep learningframeworkandalgorithms forsurvivalratepredictionoflungcancersubtypeswithregionofinterest using histopathology imagery.\" IEEE Access 12 (2024): 155309-155329. [6] Mahum, Rabbia, and Abdulmalik S. Al-Salman. \"Lung-RetinaNet: Lung cancer detectionusingaRetinaNetwithmulti-scalefeaturefusionandcontextmodule.\"IEEEAccess11(2023):53850-53861. [7] Raja,Gunasekaran,etal.\"Synergisticanalysisoflungcancer\'simpactonCardiovascularDisease using ML-based techniques.\" IEEE Journal of Biomedical and Health Informatics(2024). [8] Charan,KurupatiSai,etal.\"TransferLearningBasedMulti-ClassLungDiseasePredictionUsingTexturalFeaturesDerivedFromFusionData.\"IEEEAccess12(2024):108248-108262. [9] Chowdary, G. Jignesh, and Vivek Kanhangad. \"A dual-branch network for diagnosis ofthoraxdiseasesfromchestX-rays.\"IEEEJournalof BiomedicalandHealthInformatics26.12(2022): 6081-6092. [10] Wang,Xinggang,etal.\"Aweakly-supervisedframeworkforCOVID-19classificationandlesion localization from chest CT.\" IEEE transactions on medical imaging 39.8 (2020):2615-2625. [11] Liu,Kehong.\"Stbi-yolo:Areal-timeobjectdetectionmethodforlungnodulerecognition.\"IEEe Access 10 (2022): 75385-75394. [12] Goyal,Shimpy,andRajivSingh.\"DetectionandclassificationoflungdiseasesforpneumoniaandCovid-19usingmachineanddeeplearning techniques. \"Journalof Ambient Intelligence and Humanized Computing 14.4 (2023): 3239-3259.

Copyright

Copyright © 2026 Mrs. Ashwini Bhavsar, Atharv Borhade, Aniket Gawali, Ashish Saini. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.