Deep Learning for Histopathological Image Classification: A Survey with Case Study on Lung and Colon Cancer Detection Using Transfer Learning

Abstract

This paper presents a comprehensive survey on the application of deep learning (DL) in computational pathology for classification of lung and colon cancer from histopathologi-cal images. Though manual diagnosis is paramount for patient outcomes, it remains a tedious, subjective process that is suscep-tible to inter-observer variability. This work compares founda-tional CNN architectures against the most recent SOTA mod-els and performs a reproducible base case study using transfer learning on the publicly available LC25000 dataset, consisting of 25,000 images across five classes. Our case study compares three pre-trained models-VGG16, Xception, and DenseNet121-all of which have a frozen base with a custom classifier head. The experimental results showed that DenseNet121 was the better baseline model, with an accuracy of 97.64%, outperforming the other two: Xception with 96.48% and VGG16 with 95.08%. However, this work argues that, while this baseline is strong, ba-sic classification accuracy on this benchmark is a largely solved problem, since SOTA models including Vision Transformers and hybrid networks have shown 99.8-100% accuracy. Thus, the key research frontiers have shifted. The survey bridges the gap between theoretical foundations and practical implementation, concluding that advanced XAI frameworks, ensemble methods, and federated learning should be a priority in future research work in order to tackle data pri-vacy and make the successful clinical translation of such power-ful diagnostic tools possible.

Introduction

Cancer—especially lung and colon cancer—remains a major global health burden, and early diagnosis through histopathological analysis is critical for effective treatment. Traditionally, pathologists examine H&E-stained tissue slides manually, but this process is slow, labor-intensive, and subject to human variability. Increasing screening demands and shortages of experts further strain healthcare systems, creating the need for automated diagnostic support.

Deep learning, particularly Convolutional Neural Networks (CNNs), has significantly advanced computational pathology by enabling automatic feature extraction from histopathological images. Transfer learning is especially important in this domain because medical datasets are often limited; pre-trained models (e.g., on ImageNet) can be adapted to cancer classification tasks using either feature extraction or fine-tuning.

The paper focuses on automated multi-class classification of lung and colon cancer using the LC25000 dataset, which contains 25,000 images across five categories. It evaluates foundational CNN models such as VGG16, DenseNet121, and Xception, which differ in architecture design, efficiency, and parameter usage.

Beyond achieving high classification accuracy, the survey highlights key remaining challenges for real-world clinical adoption: improving model interpretability (explainable AI), ensuring robustness across different labs and imaging conditions (generalization), and addressing data privacy and dataset scarcity.

Overall, the paper reviews deep learning progress in histopathology, compares major CNN architectures, and identifies future research directions needed to move from high-performing models to clinically reliable cancer diagnostic systems.

Conclusion

The following survey has charted the rapid and decisive pro-gression of deep learning approaches concerned with lung and colon cancer histopathological classification. This is done using the LC25000 dataset as a common analytical ground. We start by establishing a strong performance baseline through the detailed case study. It is shown that a founda-tional architecture, DenseNet121, can attain a high test accu-racy of 97.64% using a simple, reproducible transfer learning methodology. This superior performance is not arbitrary; the dense connectivity and feature-reuse mechanism inherent to DenseNet are well-matched in theory to the complex, multi-scale, and hierarchical nature of histopathological data. This survey then placed this baseline in context, given the state-of-the-art, and showed that advanced architectures, es-pecially Vision Transformers (e.g., Swin Transformer V2), have driven performance on this benchmark to 99.9–100%. This convergence effectively signifies that the problem of pure classification on this clean, balanced dataset is a “solved” problem. This “post-accuracy” era did not terminate research, but rather shifted the field’s focus from marginal gains in accu-racy to the deeper challenges standing in the way of real-world clinical deployment. Probably the biggest research frontiers lie in three related challenges: 1) Trust & Interpretability: Beyond black-box predic-tors to completely interpretable models using Explain-able AI (XAI) that allow clinicians to comprehend and verify model reasoning. Item Generalization: Dealing with variation in staining–the major source of domain shift in pathology–so that the models deployed at dif-ferent hospitals using varied staining and scanning pro-tocols remain robust. 2) Scalability & Privacy: Completely remove the possi-ble silos of data created by patient privacy regulations using Federated Learning to train models collabora-tively without exposing sensitive information. As this survey has discussed, the latter two challenges are fundamentally intertwined. Stain variability is the principal cause of the statistical heterogeneity that often leads stan-dard FL algorithms to fail. Thus, the future of computational pathology is not about achieving another 0.01% increase on a static benchmark but about building robust, explainable, and privacy-preserving AI systems that can learn collabora-tively and generalize effectively across the diverse and dy-namic data encountered in real-world clinical practice.

References

[1] H. Sung et al., “Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries,” CA: A Cancer Journal for Clinicians, vol. 71, no. 3, pp. 209–249, 2021. [2] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436–444, 2015. [3] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet classification with deep convo-lutional neural networks,” in Proc. Advances in Neural Information Processing Systems (NIPS), 2012, pp. 1097–1105. [4] J. Deng et al., “ImageNet: A large-scale hierarchical image database,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2009, pp. 248–255. [5] A. A. Borkowski et al., “Lung and Colon Cancer Histopathological Image Dataset (LC25000),” arXiv:1912.12142, 2019. [6] R. Guidotti et al., “A survey of methods for explaining black box models,” ACM Comput. Surveys, vol. 51, no. 5, pp. 1–42, 2018. [7] D. Tellez et al., “Quantifying the effects of data augmentation and stain color normalization in convolutional neural networks for computational pathology,” Med. Image Anal., vol. 58, p. 101544, 2019. [8] M. Ilse, J. M. Tomczak, and M. Welling, “Attention-based deep multiple instance learning,” in Proc. ICML, 2018, pp. 2127–2136. [9] N. Tajbakhsh et al., “Convolutional neural networks for medical image analysis: Full training or fine tuning?,” IEEE Trans. Med. Imaging, vol. 35, no. 5, pp. 1299–1312, 2016. [10] Y. Gal and Z. Ghahramani, “Dropout as a Bayesian approximation: Representing model uncertainty in deep learning,” in Proc. ICML, 2016, pp. 1050–1059. [11] J. Yosinski et al., “How transferable are features in deep neural networks?,” in Proc. NeurIPS, 2014, pp. 3320–3328. [12] S. Abd El-Ghany et al., “Robustness fine-tuning deep learning model for cancers diagnosis based on histopathology images,” Scientific Reports, vol. 13, no. 1, p. 19572, 2023. [13] M. M. M. Hassan et al., “An Advanced Deep Learning Fusion Model for Multi-Classification of Lung and Colon Cancers Using Histopathological Images,” Diagnostics, vol. 14, no. 20, p. 2274, 2024. [14] P. Sudhakar et al., “Exploring Explainable AI Techniques for Improved Interpretability in Lung and Colon Cancer Classification,” ResearchGate, 2025. [15] V. Dabass et al., “Automated Lung and Colon Cancer Classification Using Histopathological Images,” bioRxiv, 2024. [16] J. Ji et al., “Automated lung and colon cancer classification using medical imaging based on Swin Transformer V2,” Frontiers in Oncology, vol. 14, 2024. [17] G. Huang et al., “Densely Connected Convolutional Networks,” in Proc. CVPR, 2017, pp. 2261–2269. [18] S. Suara et al., “Is Grad-CAM explainable in medical images?,” arXiv:2307.10506, 2023. [19] M. Hägele et al., “Resolving challenges in deep learning-based analyses of histopathologi-cal images using explanation methods,” Scientific Reports, vol. 10, p. 16901, 2020. [20] H. Chen et al., “HIPPO: A framework for Histopathology Interventions of Patches for Predic-tive Outcomes in computational pathology,” bioRxiv, 2024. [21] B. McMahan et al., “Communication-efficient learning of deep networks from decentralized data,” in Proc. AISTATS, 2017, pp. 1273–1282. [22] M. J. Sheller et al., “Federated learning in medicine: Facilitating multi-institutional collabo-rations without sharing patient data,” Scientific Reports, vol. 10, p. 12598, 2020. [23] M. Macenko et al., “A method for normalizing histology slides for quantitative analysis,” in Proc. ISBI, 2009, pp. 1107–1110. [24] M. Asadi-Aghbolaghi et al., “Learning generalizable AI models for multi-center histopathol-ogy classification,” Medical Image Analysis, vol. 91, p. 103038, 2024. [25] N. Coudray et al., “Classification and mutation prediction from non-small cell lung cancer histopathology images using deep learning,” Nature Medicine, vol. 24, no. 10, pp. 1559–1567, 2018.

Copyright

Copyright © 2026 Ganesh Magar, Sanket Kakade, Pranav Kolte, Yogesh Handge. 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.