Fig. 3: Token-level decoding framework and performance evaluation.

a Explicit context augmentation strategy designed to incorporate contextual information by combining tokens into token samples. b Model training pipeline: the teacher model is pre-trained on healthy samples, then fine-tuned on patient samples; knowledge distillation transfers learned features to a student model for efficient prediction. c Comparison of decoding accuracy across different numbers of tokens per sample, showing optimal performance when sufficient contextual information is included. d Accuracy improvement with word repetition in transfer learning process, demonstrating a jump from zero-shot inference (43.3%) to few-shot learning (92.2%) as repetitions increase. e Comparison of model performance across architectures with varying accuracy, FLOPs, and parameter counts; ResNet-101 and ResNet-18 were selected as the teacher and student models, respectively. f Confusion matrix for the final student model. g UMAP visualization of extracted features from the student model, illustrating token clustering patterns that indicate effective decoding and clear separation of different classes.