Johnyquest7/ResNet18

Thyroid Nodule Classification - ResNet-18 (PEMV-Style Correct)

Trained on TN5000 with exact PEMV paper recipe, optimized for AUC-ROC.

Key Recipe Differences from Failed Runs

• No ImageNet normalization (only ToTensor to [0,1])
• CrossEntropyLoss with 2 logits (not BCE with 1 logit)
• ResNet-18 (proven 85.68% accuracy baseline on TN5000)
• AdamW lr=1e-4, wd=0.05, batch=16, 128x128
• Constant LR for 200 epochs (no scheduler)

Test Set Performance

Metric	Value	95% CI
Accuracy	0.8891	-
Sensitivity	0.9175	[0.8948, 0.9366]
Specificity	0.8182	[0.7685, 0.8611]
PPV	0.9266	[0.9048, 0.9447]
NPV	0.7986	[0.7481, 0.8430]
AUC-ROC	0.9313	[0.9125, 0.9483]

References

• PEMV-Thyroid (arXiv:2603.28315): Prototype-Enhanced Multi-View Learning for Thyroid Nodule Ultrasound Classification

Generated by ML Intern

This model repository was generated by ML Intern, an agent for machine learning research and development on the Hugging Face Hub.

• Try ML Intern: https://smolagents-ml-intern.hf.space
• Source code: https://github.com/huggingface/ml-intern

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = 'Johnyquest7/Thyroid_EfficientNetV2'
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

For non-causal architectures, replace AutoModelForCausalLM with the appropriate AutoModel class.