Optimal Adaptation Depth for Medical Image Classification: Comparing Convolutional and Transformer Architectures Across Fine-Tuning Strategies

Table: Mean test metrics (Acc, AUC, F1) across fine-tuning strategies by dataset. Values are aggregated for the selected backbone; top-performing strategies within each dataset-metric column are highlighted.

Figure: Dataset-wise trends across fine-tuning strategies for DenseNet-121 and Swin-Tiny. Points show mean test performance for each strategy, with datasets encoded by color. Backbone is encoded by mark style (DenseNet-121: circles with solid lines; Swin-Tiny: squares with dashed lines).

Model Weights

Direct download links for each dataset and backbone from the Hugging Face repository.