TL;DR
A hybrid architecture in which ResNet50 is employed for localized spatial feature extraction, while Vision Transformer enables global contextual learning to automatically classify kidney tumors into multiple classes is proposed.
Renal pathology represents a diverse set of diseases that present significant clinical relevance. Included among the various types of renal pathologies are renal stones, cysts, and renal malignancies, all of which require diagnosis and therapy to prevent progression of the disease process. The current research study was performed to create and validate a classification model based on deep learning using a convolutional neural networks (CNN) architecture, namely a 50-layer Residual Network (ResNet-50) using Gradient-weighted Class Activation Mapping (Grad-CAM), to provide improved automatic detection of renal pathology from medical images and improve the interpretability of those medical images. During the study, the Explainable Deep Learning Pipeline (X-DLP) paradigm was followed, which provides a structured methodology to perform research with the use of deep learning in medical imaging. The X-DLP structures the research process into a series of phases, including Data acquisition and curation, Preprocessing and Augmentation, Model Creation via Transfer Learning, and lastly, Interpretability and Visualization.The results obtained show that the proposed model performs consistently well across different evaluation metrics. The Precision–Recall curve, with a PR-AUC close to 0.89, suggests that the model is effective at identifying positive cases even when the data are imbalanced. In addition, the F1-score reaches a peak of around 0.835 at a threshold near 0.45, indicating a good trade-off between precision and recall. From another perspective, the evaluation using Youden’s criterion reveals sensitivity and specificity values close to 0.80, which supports the model’s ability to distinguish between classes with reasonable accuracy. Moreover, the lift and cumulative gain analysis further highlight its practical usefulness, with a lift of 3.5 in the top 10% and a cumulative gain of 75% when considering 30% of the population. These results indicate that the model can effectively prioritize the most relevant positive cases. Overall, these findings suggest that the model can serve as a valuable support tool in medical diagnosis. By enabling automated classification of renal images and providing visual insights through interpretability techniques, it helps streamline clinical decision-making, reduces reliance on purely manual assessments, and enhances its potential for real-world application.
Early and accurate detection of Lung Cancer remains a major challenge in medical imaging. This paper investigates deep learning approaches for the classification of lung CT images into three categories: normal, benign, and malignant. Three convolutional neural network architectures are evaluated, namely a custom CNN, MobileNetV2, and Xception. The proposed workflow relies on a two-stage data strategy. The models are first trained on the LIDC-IDRI dataset to learn general lung and nodule-related features, and then fine-tuned on IQ-OTH/NCCD for the target three-class classification task. A unified preprocessing pipeline is applied, including resizing, rescaling, label encoding, batching, and shuffling. Performance is assessed using accuracy, loss, recall, and F1-score, with an additional 5-fold cross-validation experiment to evaluate robustness. The results show that Xception achieved the best predictive performance, with an accuracy of 0.993 and an F1-score of 0.992. MobileNetV2 also delivered strong results, reaching an accuracy of 0.982 and an F1-score of 0.981, while offering lower computational cost. The custom CNN remained competitive, with an accuracy of 0.961 and an F1-score of 0.930. The cross-validation results further supported the effectiveness of the workflow, highlighting a clear performance–efficiency trade-off between the evaluated models. The source code is publicly available on LungNet-CT.
Introduction Kidney-related disorders are one of the global health concerns that require timely detection to prevent severe health complications. The use of computed tomography (CT) images for accurate classification of kidney diseases is important. However, it is challenging to differentiate between classes due to the subtle visual differences. This study introduces a novel two-stage deep learning architecture that integrates self-supervised representation learning with supervised classification for kidney CT image analysis using a publicly available kidney CT image dataset. Methods In the first stage, the DINO framework with a Data-efficient Image Transformer (DeiT-Tiny) backbone is used to learn useful features from kidney CT images independent of labels. In the second stage, the pre-trained model is fine-tuned using labeled data to classify kidney abnormalities. To ensure model transparency and clinical trustworthiness, two explainable AI techniques are applied. Grad-CAM++ is used to highlight important regions contributing to predictions in kidney CT images. In addition, DINO’s inherent multi-head self-attention mechanism is analyzed across all attention heads to capture diverse attention patterns. Results and discussion Experimental findings indicate that the proposed framework achieves strong classification performance, with a test accuracy of 99.16%, AUC-ROC of 99.99%, F1 score of 98.97%, precision of 98.90%, and recall of 99.05%, while also providing clear interpretability for automated kidney disease classification. External validation on a CT dataset from Iraq has yielded 97.03% accuracy, supporting the generalizability of the proposed framework.
Lung cancer is the main cause of death related to cancer globally, taking the lives of about 1.8 million people each year. Detecting it early and accurately from CT scan images is very important for better patient results. Three sophisticated Convolutional Neural Network (CNN) models—ResNet50, DenseNet121, and EfficientNetB0—are used in this comprehensive study to automatically classify lung CT scan images into two categories (benign and malignant). The IQ-OTH/NCCD Lung Cancer Dataset, which is openly accessible on Kaggle, was used to train and evaluate the models. This dataset was created by Aditya Mahimkar and contains 1300 CT scan slices from 110 different patient cases. After preprocessing, only benign and malignant CT scan images were used for binary classification. To evaluate the performance of each model following training through transfer learning and fine-tuning, metrics such as accuracy, precision, recall, F1-score, and the confusion matrix were employed. ResNet50 outperformed DenseNet121 and EfficientNetB0 in terms of accuracy. A combined Weighted Ensemble Model that uses the probability outputs from all three networks was also created, which improved the overall classification performance and how well the model works in different situations. This study shows that using deep learning and combining models can be very helpful tools in detecting lung cancer during clinical screenings.
Magnetic resonance imaging is widely used for the examination of brain abnormalities because it provides detailed soft-tissue information without ionizing radiation. Nevertheless, manual interpretation of large numbers of MRI slices is time-consuming and may be affected by inter-observer variation. This paper presents a structured deep learning framework for multi-class brain tumor classification using the EfficientNet family of convolutional neural networks. The framework emphasizes consistent preprocessing, transfer learning, class-balanced augmentation, careful validation, and clinically meaningful performance reporting. Rather than treating classification as an isolated model-training task, the proposed approach connects data quality, model calibration, error analysis, and reproducibility. The paper also explains why EfficientNet is suitable for medical image classification: its compound scaling strategy balances network depth, width, and input resolution, enabling strong feature learning with comparatively efficient computation. A complete experimental protocol is described for separating training, validation, and test data at the patient level; controlling information leakage; selecting evaluation metrics; and comparing EfficientNet variants with conventional convolutional baselines. The resulting framework can support reliable thesis-level experimentation and can later be extended with visual explanation methods such as Grad-CAM. The study concludes that model efficiency alone is insufficient; dependable brain tumor classification requires disciplined data handling, transparent reporting, and external validation.