# Grad-CAM

8 related articles

Subscribe to updates

medRxiv Sep 27, 2026
Fetal ultrasound screening learns from scarce labels: RadarMAE's radar-style masking holds cross-hospital metric drops to 0.04
First-trimester ultrasound images for Down syndrome screening are low-resolution and structurally variable, and confirmed rare cases are scarce, so standard self-supervised pretraining struggles. RadarMAE instead masks alternating concentric rings and angular sectors, a radar-like pattern that forces the network to learn the continuous curvature and elongated shapes of craniofacial structures, reaching AUC 0.99 and 97% accuracy on an internal cohort. With domain-adversarial adaptation, the model calibrates on unlabeled images from four external hospitals and limits the drop in key metrics to 0.04 or less.
First-trimester ultrasound images for Down syndrome screening are low-resolution and structurally variable, and confirmed rare cases are scarce, so standard self-supervised pretraining struggles. RadarMAE instead masks alternating concentric rings and angular sectors, a radar-like pattern that forces the network to learn the continuous curvature and elongated shapes of craniofacial structures, reaching AUC 0.99 and 97% accuracy on an internal cohort. With domain-adversarial adaptation, the model calibrates on unlabeled images from four external hospitals and limits the drop in key metrics to 0.04 or less.
West China Second University Hospital of Sichuan University Central South University Sichuan University Fudan University Capital Medical University Chinese Academy of Sciences Third Affiliated Hospital of Zhengzhou University Beijing Obstetrics and Gynecology Hospital
Yantong Zhu（First author）Hua Chen（Corresponding author）Qingqing Wu（Corresponding author）
RadarMAE Domain Adaptation Fetal Ultrasound Analysis Masked Autoencoder Grad-CAM
Communications Materials Oct 19, 2026
Diffractograms are not photos: an ImageNet-only model reads phase fractions from background noise, while physics pre-training locks onto Bragg rings
Synchrotron X-ray diffraction produces a 2D diffractogram per second, and image models seem able to read α and β phase fractions of Ti–6Al–4V from them. Two models reach similar accuracy after fine-tuning on 250 experimental diffractograms, yet robustness tests and attention maps show the ImageNet-only model infers fractions from background and pixel noise, while the model pre-trained on simulated diffractograms relies on Bragg reflections. Accuracy alone does not validate a model.
Synchrotron X-ray diffraction produces a 2D diffractogram per second, and image models seem able to read α and β phase fractions of Ti–6Al–4V from them. Two models reach similar accuracy after fine-tuning on 250 experimental diffractograms, yet robustness tests and attention maps show the ImageNet-only model infers fractions from background and pixel noise, while the model pre-trained on simulated diffractograms relies on Bragg reflections. Accuracy alone does not validate a model.
Case Western Reserve University
Mohommad Redad Mehdi（First author）Frank Ernst（Corresponding author）
Transfer Learning Surface-Texture Bias Ti–6Al–4V Grad-CAM
arXiv Oct 12, 2026
Where a model looks is not where people look: gaze heatmaps as a benchmark show Mamba localizes chest X-rays well but misses human intensity
A cross-architecture comparison of CNNs, ViTs and Vision Mamba uses human eye-tracking heatmaps as an independent benchmark to quantify how closely model saliency matches human gaze, finding that Mamba reaches the highest localization scores on chest X-rays while its saliency intensity is nearly uncorrelated with human fixations.
Marco Stefani（First author）
Explainable AI Saliency Maps Grad-CAM Vision Mamba Vision Transformers
researchsquare.com Oct 14, 2026
Paddy pest ID now ships with a reason and a tamper-proof log: 97.83% accuracy, 34.2% less pesticide
AgriMind-PestGuard couples three parts for paddy pest management: a multi-scale attention CNN that identifies 14 major pest species, a Grad-CAM-HCA module that grounds each decision in entomological concepts, and a Hyperledger Fabric layer that writes detections, pesticide recommendations and compliance records into a tamper-proof ledger. On a curated 58,400-image field dataset it reaches 97.83% classification accuracy, above the strongest baseline EfficientNetV2-S at 96.38%; explanation faithfulness is 0.91 and three entomologists agree with 93.7% of concept attributions. Across 12 paddy farms, field trials report 34.2% less pesticide use and 8.7% higher yield than conventional practice.
AgriMind-PestGuard couples three parts for paddy pest management: a multi-scale attention CNN that identifies 14 major pest species, a Grad-CAM-HCA module that grounds each decision in entomological concepts, and a Hyperledger Fabric layer that writes detections, pesticide recommendations and compliance records into a tamper-proof ledger. On a curated 58,400-image field dataset it reaches 97.83% classification accuracy, above the strongest baseline EfficientNetV2-S at 96.38%; explanation faithfulness is 0.91 and three entomologists agree with 93.7% of concept attributions. Across 12 paddy farms, field trials report 34.2% less pesticide use and 8.7% higher yield than conventional practice.
Hindustan Institute of Technology and Science
Rambabu Kambhampati（First author）
AgriMind-PestGuard Multi-Scale Attention-augmented CNN Pest Classification Grad-CAM Hyperledger Fabric
Journal of Bone and Joint Surgery Sep 28, 2026
A deep learning model using pre-reduction radiographs and clinical-demographic data predicted hand surgeons' surgical recommendations for distal radial fractures, reaching 87.14% accuracy and 97% sensitivity on the test set
This feasibility study trained a convolutional neural network on pre-reduction injury radiographs and combined its outputs with clinical and demographic data in a random forest model to predict whether a group of fellowship-trained hand surgeons at one institution would recommend operative intervention for distal radial fractures in 1,040 patients (884 training, 156 testing); on the test set the combined model achieved 87.14% accuracy, 97% sensitivity, 73% specificity, an area under the ROC curve of 0.96, and a Brier score of 0.10, with Grad-CAM indicating the CNN focused on clinically relevant features such as fracture displacement and SHAP highlighting age and lateral wrist radiographs as key contributors.
This feasibility study trained a convolutional neural network on pre-reduction injury radiographs and combined its outputs with clinical and demographic data in a random forest model to predict whether a group of fellowship-trained hand surgeons at one institution would recommend operative intervention for distal radial fractures in 1,040 patients (884 training, 156 testing); on the test set the combined model achieved 87.14% accuracy, 97% sensitivity, 73% specificity, an area under the ROC curve of 0.96, and a Brier score of 0.10, with Grad-CAM indicating the CNN focused on clinically relevant features such as fracture displacement and SHAP highlighting age and lateral wrist radiographs as key contributors.
University of Pennsylvania Hospital for Special Surgery
Omar Shareef（First author）
Convolutional Neural Network Random Forest Grad-CAM SHAP Distal radial fracture operative recommendation prediction
Journal of Zhejiang University(Science Edition) Sep 15, 2026
An Explainable Multimodal Deep Learning Framework for Alzheimer's Disease Classification Using MRI, PET, and Clinical Data
Using ADNI data, this study builds a patient-level multimodal fusion framework based on ResNet50 transfer learning that concatenates MRI and PET imaging features with clinical variables (ADAS11, ADAS13, APOE4, age, sex, education, MMSE total) in a fully connected network to produce three-way AD/MCI/CN predictions, with Grad-CAM heatmaps for explanation; the ablation shows accuracy rising as modalities are added, from 39.61% for MRI alone and 51.72% for PET alone to 66.67% for clinical-only, 54.17% for MRI+PET, and 79.17% for the full three-modality model on a 24-patient test cohort (95% Wilson interval roughly 59.5%-90.8%), with the addition of clinical data producing the single largest gain.
Using ADNI data, this study builds a patient-level multimodal fusion framework based on ResNet50 transfer learning that concatenates MRI and PET imaging features with clinical variables (ADAS11, ADAS13, APOE4, age, sex, education, MMSE total) in a fully connected network to produce three-way AD/MCI/CN predictions, with Grad-CAM heatmaps for explanation; the ablation shows accuracy rising as modalities are added, from 39.61% for MRI alone and 51.72% for PET alone to 66.67% for clinical-only, 54.17% for MRI+PET, and 79.17% for the full three-modality model on a 24-patient test cohort (95% Wilson interval roughly 59.5%-90.8%), with the addition of clinical data producing the single largest gain.
Mathrusri Ramabai Ambedkar Dental College & Hospital
Dr. K R Shylaja（First author）
Alzheimer's Disease Diagnosis Multimodal Fusion Grad-CAM Transfer Learning
Primary Health Care Research & Development Sep 15, 2026
A Reliability-Oriented Hybrid Deep Learning Framework for Early Autism Screening
This study proposes a hybrid deep learning framework that fuses a static periocular ResNet18 classifier with a facial-image ensemble of ResNet50, EfficientNet-B0 and DenseNet121 under an OR-based parallel rule for early autism spectrum disorder risk indication and referral support; the periocular model reached 90% sensitivity, the facial ensemble reached 87.1% sensitivity with an AUC of 0.948, and under a conditional-independence assumption the analytically estimated system-level sensitivity was 0.9871, corresponding to a joint false-negative probability of about 1.29%, while system specificity fell to roughly 0.807.
This study proposes a hybrid deep learning framework that fuses a static periocular ResNet18 classifier with a facial-image ensemble of ResNet50, EfficientNet-B0 and DenseNet121 under an OR-based parallel rule for early autism spectrum disorder risk indication and referral support; the periocular model reached 90% sensitivity, the facial ensemble reached 87.1% sensitivity with an AUC of 0.948, and under a conditional-independence assumption the analytically estimated system-level sensitivity was 0.9871, corresponding to a joint false-negative probability of about 1.29%, while system specificity fell to roughly 0.807.
Mudanya Üniversitesi
Hüseyin Ünözkan（First author, Corresponding author）
Hybrid Deep Learning Framework Transfer Learning Ensemble Learning Grad-CAM Autism Screening
Scientific Reports Sep 14, 2026
Scaling vision models does not consistently improve localisation-based explanation quality
This study evaluated 11 vision models from the ResNet, DenseNet, and Vision Transformer families (seven trained from scratch, four pretrained) on the Oxford-IIIT Pet and Chest X-ray Pneumothorax datasets with ground-truth segmentation masks, generated explanations with five post-hoc XAI methods (Saliency, GradientSHAP, Integrated Gradients, Feature Permutation, Grad-CAM), and quantified mask alignment using Relevance Rank Accuracy and the proposed Dual-Polarity Precision; it found that increasing depth and parameter count did not improve explanation quality in most statistical comparisons, that smaller models often matched or exceeded deeper variants, that pretraining typically improved predictive performance and increased the dependence of explanations on learned weights without consisten
This study evaluated 11 vision models from the ResNet, DenseNet, and Vision Transformer families (seven trained from scratch, four pretrained) on the Oxford-IIIT Pet and Chest X-ray Pneumothorax datasets with ground-truth segmentation masks, generated explanations with five post-hoc XAI methods (Saliency, GradientSHAP, Integrated Gradients, Feature Permutation, Grad-CAM), and quantified mask alignment using Relevance Rank Accuracy and the proposed Dual-Polarity Precision; it found that increasing depth and parameter count did not improve explanation quality in most statistical comparisons, that smaller models often matched or exceeded deeper variants, that pretraining typically improved predictive performance and increased the dependence of explanations on learned weights without consisten
University of Antwerp University of Warsaw
Mateusz Cedro（First author）Marcin Chlebus（Corresponding author）
Dual-Polarity Precision Explainable AI Feature attribution Saliency Maps Grad-CAM

You’re all caught up