The research, conducted by Elham Tahsin Yasin, Ilkay Cinar, and Murat Koklu at Selcuk University in Turkey, set out to benchmark five well-known convolutional neural network architectures on mammographic image classification: DenseNet201, InceptionResNetV2, MobileNetV2, NASNetLarge, and Xception. Rather than building networks from scratch, the team used transfer learning, a strategy in which models pre-trained on millions of everyday images are repurposed for a specialized task. The underlying idea is that features learned from natural photographs, such as edges, textures, and shapes, provide a useful starting point that can be fine-tuned for the specific visual patterns of cancerous versus non-cancerous breast tissue.

The experimental design was deliberately comparative. Each of the five architectures was trained and evaluated under two distinct conditions: once using the original dataset of 745 mammographic images, and once using an expanded dataset of 9,685 images produced through data augmentation. Augmentation artificially enlarges a training set by applying transformations such as rotations, flips, and other image manipulations that preserve the diagnostic content of each image while creating new variations for the network to learn from. This matters enormously in medical imaging, where collecting thousands of annotated patient images is expensive, slow, and constrained by privacy regulations. The question the researchers posed was straightforward: how much does augmentation actually help when the task is distinguishing cancerous from non-cancerous mammograms?

The answer, it turns out, is quite a lot. Without augmentation, DenseNet201 led the field with an accuracy of 96.24 percent, followed closely by InceptionResNetV2 and Xception, both at 95.43 percent, NASNetLarge at 95.31 percent, and MobileNetV2 at 94.64 percent. Even in this baseline condition, the models demonstrated that transfer learning can extract clinically relevant information from a relatively small dataset of just 745 images. But the numbers climbed across the board once augmentation was applied to the larger dataset. DenseNet201 again took first place, improving to 98.06 percent accuracy, while MobileNetV2 reached 97.76 percent, Xception 97.67 percent, InceptionResNetV2 97.63 percent, and NASNetLarge 97.55 percent.

Several aspects of these results deserve attention. First, the gap between the best and worst models narrowed dramatically after augmentation, from roughly 1.6 percentage points to just half a point. This suggests that with sufficient training data, the choice of architecture becomes less critical than the quality and quantity of the training examples, a finding with practical implications for research groups with limited computational budgets. Second, MobileNetV2, the lightest and most computationally efficient model in the lineup, closed most of the gap with its heavier rivals, hinting that lightweight networks suitable for deployment in clinics with modest hardware could perform nearly as well as resource-hungry giants.

DenseNet201’s consistent dominance is also technically interesting. The DenseNet family of architectures is built around dense connectivity, in which each layer receives feature maps from all preceding layers. This design encourages feature reuse, strengthens gradient flow during training, and tends to perform well with limited data, which may explain why it excelled in both experimental conditions. Xception, by contrast, relies on depthwise separable convolutions that separate the processing of spatial patterns and cross-channel relationships, while NASNetLarge was discovered through automated neural architecture search rather than human design. The fact that all three approaches converge on similar performance after augmentation underscores a broader trend in the field: modern architectures have largely saturated the achievable accuracy on well-defined binary classification tasks, leaving data quality as the main lever for improvement.

The study is also notable for its explainability-aware framing. In medical applications, a model that simply outputs a percentage score is of limited use if clinicians cannot see why it reached its decision. Techniques such as gradient-weighted class activation mapping, which highlight the image regions most influential in a network’s prediction, have become central to building trust in artificial intelligence diagnostics. By benchmarking models with explainability in mind, the researchers align their work with a growing consensus that accuracy alone is not enough for clinical adoption; a model must also offer interpretable evidence that its attention falls on the actual lesion rather than on artifacts, imaging markers, or background tissue.

The dataset underpinning the study, known as Mammogram Mastery, was made publicly available through the Mendeley Data repository, which is significant for reproducibility. Medical imaging research has long been hampered by restricted access to patient data, and open datasets allow independent teams to verify results, compare methods on equal footing, and build cumulative knowledge. The researchers also employed stratified k-fold cross-validation, a rigorous evaluation strategy that preserves the class balance of cancerous and non-cancerous images across each validation split, reducing the risk that reported accuracies are inflated by lucky partitions of the data.

What do these numbers mean for patients? An accuracy of 98 percent on a binary classification task is impressive, but the path from benchmark to bedside involves many additional hurdles. Clinical systems must handle the full complexity of real screening practice, including different imaging devices, varying breast densities, image quality issues, and the far more granular classification schemes radiologists actually use, such as the BI-RADS categories. False positives generate unnecessary anxiety and biopsies, while false negatives can delay life-saving treatment, so the balance of sensitivity and specificity matters as much as raw accuracy. The authors’ results should therefore be read as strong evidence of feasibility rather than a claim of clinical readiness.

Nevertheless, the trajectory is clear and encouraging. The study demonstrates that data augmentation can lift the performance of pre-trained deep learning models by more than a full percentage point even at very high baseline accuracy, and that the resulting systems are accurate enough to serve as reliable second readers or triage tools that flag suspicious mammograms for priority review. As explainable artificial intelligence methods mature and open datasets grow, the combination of transfer learning, augmentation, and transparent decision-making may bring automated breast cancer detection from the research laboratory into routine screening practice, helping radiologists catch cancers earlier, when treatment is most effective.

Subject of Research: Explainability-aware benchmarking of pre-trained deep learning models for breast cancer classification in mammographic images

Article Title: Explainability-aware benchmarking of pre-trained deep learning models for breast cancer classification in mammographic images

Article References: Yasin, E. T., Cinar, I., & Koklu, M. (2026). Explainability-aware benchmarking of pre-trained deep learning models for breast cancer classification in mammographic images. Neural Computing and Applications, 38(19), Article 795. https://doi.org/10.1007/s00521-026-12513-1

Image Credits: AI Generated

DOI: 10.1007/s00521-026-12513-1

Keywords: breast cancer, mammography, deep learning, transfer learning, data augmentation, DenseNet201, convolutional neural networks, medical imaging, explainable AI, cancer screening, computer-aided diagnosis, benchmarking

The core idea behind SkinEnsemNet is deceptively simple. Instead of betting everything on one neural network, the researchers combined three pretrained convolutional neural network backbones that each see dermoscopic images through a different computational lens. The first, VGG16, is a classic deep architecture known for capturing localized spatial and hierarchical features through its stack of small convolutional filters. The second, DenseNet201, belongs to a family of densely connected networks in which each layer receives feature maps from all preceding layers, enabling extensive feature reuse and efficient hierarchical extraction. The third, EfficientNetB0, was designed to balance accuracy against computational cost through a systematic compound scaling of network depth, width, and resolution. By pooling the strengths of these three very different models, the framework aims to exploit complementary representations of the same skin lesion image, something no single architecture can fully achieve on its own.

The technical pipeline behind the system is worth unpacking, because it illustrates the careful engineering that underpins modern medical AI. The researchers began with transfer learning, meaning that each of the three backbones had already been trained on ImageNet, a massive general-purpose image database, before being adapted to dermoscopy. Transfer learning allows models to inherit rich visual features learned from millions of natural images, which is especially valuable when medical training data are limited. Before any image reached the networks, the pipeline applied normalization to standardize pixel values, and it tackled one of the most persistent problems in dermatological datasets: class imbalance. Some lesion categories are far more common than others, and naive models tend to ignore rare but clinically dangerous classes. To counter this, the team used random oversampling, duplicating examples from underrepresented classes until the training distribution was more balanced.

Once the three backbone networks had processed an image, their outputs were not simply averaged or voted on. Instead, SkinEnsemNet performs feature-level fusion. Global average pooling condenses each backbone’s high-dimensional feature maps into compact vectors, and these vectors are then concatenated into a single joint representation that carries spatial, hierarchical, and efficiency-oriented information simultaneously. On top of this fused representation sits a stack of dense classification layers, with dataset-specific heads allowing the same framework to be tailored to different benchmarks. Dropout regularization was applied throughout to prevent the network from memorizing training examples, a critical safeguard when the goal is generalization to unseen lesions. This design means the ensemble learns how to weigh and combine evidence from its three constituent models, rather than relying on a crude post-hoc vote.

The performance numbers reported in the study are impressive. On held-out image-level test partitions, SkinEnsemNet achieved an accuracy of 98.85 percent on the HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions, and 97.83 percent on ISIC-2019, a broader benchmark spanning eight lesion categories. The corresponding weighted F1-scores, which balance precision and recall while accounting for class frequencies, were 98.86 percent and 97.83 percent respectively. Crucially, the ensemble outperformed each of the individual CNN backbones when those were evaluated under the same experimental protocol, confirming that the fusion strategy, and not merely the quality of any single model, was responsible for the gains. In a field where a fraction of a percentage point can mean the difference between a missed melanoma and a timely diagnosis, such consistent improvements carry real weight.

Accuracy alone, however, is not enough for a tool that might one day influence clinical decisions. A model that performs well but cannot explain why it reached a conclusion is unlikely to earn the trust of dermatologists. Addressing this, the researchers applied Grad-CAM, a gradient-based visualization technique that highlights the image regions most influential in a given prediction. The resulting heatmaps offer qualitative insight into whether the network is attending to the actual lesion boundaries, pigmentation patterns, and structural features that clinicians consider, or whether it is being distracted by irrelevant artifacts such as rulers, hair, or dark corners of the dermoscopic frame. While the visualizations are qualitative rather than a formal proof of correctness, they represent an important step toward interpretable AI in dermatology, where black-box predictions have historically been met with justified skepticism.

The choice of datasets deserves attention as well. HAM10000, assembled by Tschandl and colleagues, has become a de facto standard for skin lesion classification research, containing over ten thousand dermatoscopic images across seven diagnostic categories including melanoma, basal cell carcinoma, and benign nevi. ISIC-2019, drawn from the International Skin Imaging Collaboration, extends the challenge to eight classes and greater image diversity. Testing on both benchmarks, with dataset-specific classification layers, demonstrates that the framework’s architecture is not overfit to the quirks of a single data collection. That said, the authors themselves are careful to frame these results as strong within-dataset performance, a distinction that matters enormously in the translation of laboratory models to real-world clinics.

Indeed, the study’s authors are notably candid about the limits of what they have shown and the work that remains. They call for further evaluation using patient-level splitting, in which all images from a single patient are kept on the same side of the train-test divide, preventing the model from effectively memorizing individuals. They also recommend testing on external datasets collected at different institutions and with different imaging equipment, a rigorous test of generalization that within-dataset splits cannot provide. Beyond that, they point toward calibration analysis, which assesses whether the model’s confidence scores actually reflect the true probability of a diagnosis, as well as formal clinician assessment and deployment-oriented experiments. These are exactly the kinds of validation steps that regulatory bodies and medical professionals demand before an algorithm can move from a research paper to a hospital workstation.

The broader context makes this work part of a fast-moving wave. Deep learning has transformed medical image analysis over the past decade, and skin cancer detection has been one of its most visible battlegrounds, with convolutional networks repeatedly matching or exceeding expert dermatologists on curated benchmarks. Yet most published systems rely on a single architecture, leaving performance vulnerable to the blind spots of that particular design. Ensemble approaches that deliberately combine heterogeneous backbones, as SkinEnsemNet does, point to a strategy that trades some computational overhead for robustness and accuracy. With EfficientNetB0 already among the most computationally efficient architectures available, the authors suggest that such a framework could eventually run in settings where resources are constrained, from primary care clinics to smartphone-based screening tools in underserved regions.

For now, SkinEnsemNet stands as a rigorous proof of concept rather than a finished clinical product, but its trajectory is clear. The framework shows that thoughtfully orchestrated architectural diversity, disciplined preprocessing, and attention to interpretability can push dermoscopic classification to near-ceiling performance on standard benchmarks. The path from there to the clinic runs through external validation, calibration, and collaboration with the dermatologists whose judgment these tools are meant to augment, not replace. If those steps bear out the promise of the current results, the humble mole photographed through a dermatoscope may soon be scrutinized not just by one pair of expert eyes, but by a committee of three very different artificial minds, each catching what the others might miss, and together edging medicine closer to the goal of catching deadly skin cancers before they ever get the chance to spread.

Subject of Research: Ensemble deep learning for skin cancer classification from dermoscopic images

Article Title: SkinEnsemNet: an ensemble deep learning-based framework for the classification of skin cancer

Article References: Lodh, E., Chowdhury, T., Majumder, S., De, M., & Singha, S. (2026). SkinEnsemNet: an ensemble deep learning-based framework for the classification of skin cancer. Multimedia Tools and Applications, 85(9), Article 726. https://doi.org/10.1007/s11042-026-21892-5

Image Credits: AI Generated

DOI: 10.1007/s11042-026-21892-5

Keywords: skin cancer, deep learning, ensemble learning, transfer learning, VGG16, DenseNet201, EfficientNetB0, HAM10000, ISIC-2019, Grad-CAM, dermoscopy, convolutional neural networks