Skip to main content
Search

# Medical Image Classification

10 related articles

Subscribe to updates
  1. arXiv

    Ultrasound labels don't need every frame: EvoDiff generates diffusion probes, then sends only retrieved real images to experts

    EvoDiff recasts active-learning sample selection as generate-then-retrieve: diffusion sampling in feature space produces synthetic probes, and only the real unlabeled ultrasound images those probes retrieve are sent to experts for annotation. Across 15 datasets and 16,696 images spanning six organ tasks, it beat random sampling and several active-learning baselines in most settings on both ResNet-50 and the USFM foundation model, and a model-based annotation-effort metric estimated a 24% to 84% workload reduction relative to random sampling.

  2. arXiv

    Output parity alone misses how predictions are built: adding an attribution-disparity penalty improves all six medical-imaging fairness trade-offs

    Fairness in medical image classification is usually enforced only on output-level group gaps, yet with a linear classification head the output score gap equals the signed sum of feature-attribution differences, so opposing contributions can cancel and make a model look fair. Adding the same attribution-disparity penalty to three equalized-odds surrogates produced positive paired hypervolume gains in all six comparisons on HAM10000 gender fairness and Harvard-GF race fairness, with relative gains of 1.49% to 10.57%. Under a predefined utility constraint, the augmented model families also cut equalized-odds difference by 9.0% to 38.7% while attribution disparity and the cancellation ratio fell.

  3. arXiv

    An LLM wrote a medical image classifier without seeing one image, but 256 labeled scans still beat it on every task

    Claude Opus 5.5 and GPT-6 Astra each wrote the weights of a small convolutional network in a single reply, without access to any image, reaching mean test AUC of 0.719 and 0.711 across six MedMNIST tasks, above zero-shot MedSigLIP at 0.588. Yet the same network trained with Adam on 256 labeled images reached a mean AUC of 0.828 and was higher on every MedMNIST task; on the synthetic uterine MRI task Opus reached 0.764 while Astra fell to 0.136 through a systematic error.

  4. Saliency maps for medical image classifiers stop guessing from gradients: unsupervised training lets a frozen classifier check the mask

    Saliency maps for convolutional classifiers on medical images now have a new training route: SalientNet needs no annotated saliency ground truth and instead requires the generated mask to keep enough information for a frozen classifier to reproduce its original prediction within a controlled tolerance. Across BreakHis, Brain MRI ND5 and FracAtlas, it reaches higher faithfulness (Win %) and lower prediction Drop (%) than Grad-CAM, Grad-CAM++ and LIME in nearly all dataset-architecture combinations, and on FracAtlas its maps correlate spatially with practitioner-annotated fracture regions, with a mean ground-truth active fraction of up to 0.96 for ResNet-18.

  5. medRxiv

    Lightweight medical CNNs let recurrence order set channel width: brain-MRI parameters fall up to 50.4%, with accuracy gains and losses

    Channel widths in a lightweight medical image classifier no longer have to be set by hand or found by architecture search. Fibonacci-Net allocated filters by a fixed second-order recurrence, leaving little control over capacity. Scaled m-BonacciNet derives a scaling coefficient from the dominant root of the m-Bonacci recurrence, narrows the original channel vector, and uses depthwise separable convolutions in the last blocks; on 15-class brain MRI, the Tribonacci variant reached 96.09% accuracy and 0.9581 macro F1 with about 0.85 million parameters and 365.69 million FLOPs, though some variants lost accuracy on 4-class brain MRI and CKD ultrasound.

  6. Attention modules add nothing on nine-class chest X-rays: five add-ons tie an identical ResNet-50 across ten seeds

    On nine-class chest radiograph classification, five lightweight attention or graph modules were each inserted into the same ResNet-50 backbone, and none beat the unmodified baseline across ten random seeds. The baseline reached a per-seed macro-AUC of 0.7990±0.0023; CBAM and GAT were significantly lower after multiple-comparison correction, yet every effect size fell inside a pre-specified 0.01-AUC equivalence margin. The result held after a 2.4-fold larger training set, on ResNet-18, EfficientNet-B0 and MobileNetV2, and under a patient-level split.

  7. Research Square

    An end-to-end agentic framework compares CNNs and vision transformers across three medical imaging tasks and adds LLM-generated structured reports

    This work presents an end-to-end comparative medical imaging framework that evaluates ResNet50, EfficientNet-B0, DenseNet121, DeiT-Small, and Swin-Tiny across three heterogeneous tasks—chest X-ray pneumonia classification, brain MRI tumor detection, and dermoscopic skin cancer classification—integrating transfer learning, class-imbalance handling, model calibration, bootstrap confidence intervals, robustness evaluation, failure-case analysis, Grad-CAM explainability, ensemble learning, and extensive performance metrics, together with an LLM-driven reporting component constrained to research-oriented assistance that generates structured model-comparison summaries, explainability interpretations, and decision-support reports; results indicate that CNNs remain highly effective for structured

  8. arXiv

    AGGRNet splits medical image features into informative and non-informative via a learnable threshold, reaching 5.48% higher accuracy than HiFuse-Small on Kvasir

    The work proposes AGGRNet, which embeds a Feature Extraction and Aggregation (FEA) module and a C2PCA block into a YOLOv11 classification backbone; spatial and channel attention plus a learnable threshold τ split feature maps into informative and non-informative parts that are then aggregated by cross-attention, yielding results above the compared SOTA models on five public datasets, including a 5.48% accuracy gain on Kvasir and a 2.2% gain on LIMUC.

  9. arXiv

    PlaneCycle lifts DINOv3's 2D weights into a 3D model with no training and no adapters, reaching 87.8 average AUC under linear probing

    The authors introduce PlaneCycle, a parameter-free, training-free, adapter-free 2D-to-3D lifting operator that lets a pretrained 2D backbone acquire 3D fusion by cyclically distributing spatial aggregation across the orthogonal HW, DW, and DH planes throughout network depth without modifying any pretrained parameter; using DINOv3 ViT-S/16, ViT-B/16, and ViT-L/16 on six 3D classification and three 3D segmentation benchmarks, PCg reaches 87.8 average AUC and 82.0 average ACC on ViT-B/16 under linear probing, surpassing slice-wise 2D and 3D-flattening baselines with paired t-test significance (p<0.05) on 5/6 datasets, and after full fine-tuning it matches standard 3D architectures, exceeding 3D flattening by up to 2.6 Dice points on segmentation while retaining 2D-level attention complexity.

  10. Research Square

    Cross-cohort evaluation of thyroid ultrasound deep learning across different outcome definitions: a duplicate-controlled benchmark

    This preprint trained five backbones under duplicate-controlled image-level partitioning (387 thyroid ultrasound images and 1,332 fine-needle aspiration cytology blocks from 385 public cases, labelled by postoperative diagnosis) and applied the frozen models to TN3K (n = 1,228, dataset-provided benign/malignant labels) and to a DDTI endpoint derived from radiologist TI-RADS categories (n = 637), finding ultrasound ensemble AUROC 0.825 and ResNet-50 0.888 on TN3K versus 0.477 and 0.437 against the TI-RADS-derived endpoint, with a cytology benchmark AUROC of 0.986 (0.970–0.998) against 0.733 for the ultrasound ensemble, while noting that cohort, acquisition and endpoint changed together so the contrast cannot be attributed to label definition alone.

You’re all caught up