# Accurate Detection of Mediastinal Lesions with nnDetection

**Authors:** Michael Baumgartner, Peter M. Full, Klaus H. Maier-Hein  
**arXiv:** [2303.11214](https://arxiv.org/abs/2303.11214)  
**Date:** 2023-03-20  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.11214v1/

## Key points

- The paper addresses the challenge of detecting mediastinal lesions, which is a rarely explored medical object detection problem.
- The authors applied a modified version of the self-configuring method nnDetection to the Mediastinal Lesion Analysis (MELA) Challenge 2022.
- Their approach involved incorporating automatically generated pseudo masks, training high capacity models with large patch sizes in a multi GPU setup, and using an adapted augmentation scheme to reduce localization errors caused by rotations.
- The authors' method achieved an excellent FROC score of 0.9922 at IoU 0.10 and 0.9880 at IoU 0.3 in their cross-validation experiments, indicating its effectiveness in accurately detecting mediastinal lesions.
- Their submitted ensemble ranked third in the competition with a FROC score of 0.9897 on the MELA challenge leaderboard.

## Summary

, , , , 
In their paper titled "Accurate Detection of Mediastinal Lesions with nnDetection," Michael Baumgartner, Peter M. Full, and Klaus H. Maier-Hein address the challenge of detecting mediastinal lesions, which is a rarely explored medical object detection problem. To tackle this issue, the authors applied a modified version of the self-configuring method nnDetection to the Mediastinal Lesion Analysis (MELA) Challenge 2022. Their approach involved incorporating automatically generated pseudo masks, training high capacity models with large patch sizes in a multi GPU setup, and using an adapted augmentation scheme to reduce localization errors caused by rotations. The authors' method achieved an excellent FROC score of 0.9922 at IoU 0.10 and 0.9880 at IoU 0.3 in their cross-validation experiments, indicating its effectiveness in accurately detecting mediastinal lesions. Furthermore, their submitted ensemble ranked third in the competition with a FROC score of 0.9897 on the MELA challenge leaderboard. This study's findings are significant as they demonstrate that applying advanced techniques such as nnDetection can improve mediastinal lesion detection accuracy significantly. The authors' approach could potentially be used to develop more accurate diagnostic tools for identifying these types of lesions early on and improving patient outcomes. Such diagnostic tools could help clinicians detect these lesions earlier and provide better treatment options for patients suffering from them.