DSDFF-YOLO Model Automates Lesion Detection on Post-Therapy Radioiodine Whole-Body Scans

A deep learning model (DSDFF-YOLO) outperformed nuclear medicine physicians in detecting thyroid remnants (sensitivity 85.53%, specificity 91.11%) and lymph node metastases (sensitivity 79.31%, specificity 93.52%) on post-therapy radioiodine whole-body scans in an independent te…

• DSDFF-YOLO incorporates depthwise-separable multi-scale fusion and dual-control frequency-domain self-attention modules.
• On the validation set of 62 patients, the model achieved mAP₅₀ of 0.772, mean precision 0.801, and recall 0.731 for detecting thyroid remnants and lymph node metastases, outperforming other state-of-the-art models.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.