Do Foundation Models Outperform Lightweight CNNs in Breast Cancer Histopathology Classification?

Arnav Dhiman

doi:10.36948/ijfmr.2026.v08i03.81631

Do Foundation Models Outperform Lightweight CNNs in Breast Cancer Histopathology Classification?

Author(s)	Mr. Arnav Dhiman
Country	India
Abstract	Accurate and computationally efficient classification of breast cancer histopathology images is critical for scalable clinical deployment of AI-assisted diagnostics. This paper benchmarks three deep learning architectures, MobileNetV3, ResNet50, and DINO Vision Transformer (ViT), on the BreakHis dataset across six dimensions: test accuracy, weighted F1-score, inference latency, memory usage, energy consumption, and model size. Using 7,909 microscopic images across four magnification factors (40X, 100X, 200X, 400X), it is shown that MobileNetV3 achieves 96.12% test accuracy with only 1.52M parameters and 0.256 ms per-image latency, while ResNet50 reaches 97.81% at the cost of 23.51M parameters and 3.331 ms latency. Contrary to expectations, DINO ViT underperforms both convolutional neural networks (91.07%) on this limited dataset, suggesting that transformer-based foundation models require either larger datasets or more extensive fine-tuning to match CNN inductive biases in medical imaging. Grad-CAM and Attention Rollout are further compared as explainability methods and an accuracy-efficiency Pareto frontier is constructed to guide model selection for clinical deployment. The findings demonstrate that lightweight CNNs offer a compelling trade-off and should be the default choice for resource-constrained pathology settings.
Keywords	Breast Cancer, Histopathology, BreakHis, MobileNetV3, ResNet50, DINO Vision Transformer, Grad-CAM, Attention Rollout, Computational Pathology, Efficiency Frontier
Field	Computer > Artificial Intelligence / Simulation / Virtual Reality
Published In	Volume 8, Issue 3, May-June 2026
Published On	2026-06-20
DOI	https://doi.org/10.36948/ijfmr.2026.v08i03.81631

View / Download PDF File

About IJFMR Fees & Payment Current Issue Publication Archive	Submit Research Paper Track Submission Status Publication Guidelines Publication Ethics Peer Review & Plagiarism	Join as a Reviewer Editors & Reviewers Reviewer Referral Program Get Reviewer Membership Certi.	Website/Journal Policies Usage Policy Content Policies Privacy Policy

Contact Us		+91-9687-828-838	editor@ijfmr.com

International Journal For Multidisciplinary Research

A Widely Indexed Open Access Peer Reviewed Multidisciplinary Bi-monthly Scholarly International Journal

Do Foundation Models Outperform Lightweight CNNs in Breast Cancer Histopathology Classification?

Share this