TABLE I. Summary of Prior Work on Wound Infection Classification Using Deep Learning.

Specific ML problem	Related Work	Summary of Approach	No. of Target Classes	Dataset	Results
Wound segmentation and Infection Classification	Wang et al. 2015 [4]	CNN-based: ConvNet + SVM	2 classes (infection and no infection)	NYU wound Database	Accuracy: 95.6% PPV: 40% Sensitivity: 31%
Classification of 7 tissue types including infection	Nejati et al. 2018 [5]	CNN-based: AlexNet + PCA + SVM		Private data (data statistics is unknown)	Accuracy 95.6% (Only reported accuracy)
DFU infection classification	Goyal et al. 2020 [6]	CNN-based: Ensemble CNN	2 classes (infection and no infection)	Part B DFU 2020 dataset (We also used this dataset)	Accuracy: 72.7% PPV: 73.5% Sensitivity: 70.9%
	Al-Garaawi et al. 2022 [7]	CNN-based: DFU-RGB-TEX-Net			Accuracy: 74.2% PPV: 74.1% Sensitivity: 75.1%
	Liu et al. 2022 [8]	CNN-based: augmentations + EfficientNet			Data leakage when splitting & performing augmentations
DFU wound ischemia and infection classification	Yap et al. 2021 [9]	CNN-based: VGG, ResNet, InceptionV3, DenseNet, EfficientNet	4 classes (both infection and ischemia, infection, ischemia, none)	DFUC2021 dataset	EfficientNet B0 performance: F1, PPV, SEN = 55%, 57%, 62%
	Galdran et al. 2021 [10]	ViT-based: ViT, DeiT, BiT			BiT performance: F1, PPV, SEN = 61%, 66%, 61%