HierAttn: Deeply Supervised Skin Lesions Diagnosis with Stage and Branch Attention

Deeply Supervised Skin Lesions Diagnosis with Stage and Branch Attention
Wei Dai, Rui Liu, Tianyi Wu, Min Wang, Jianqin Yin, Jun Liu
Accepted in IEEE JBHI, 2023. [Arxiv][Paper]

Installation

Please refer to INSTALL.md for installation instructions.

Benchmark Evaluation and Training

Please refer to DATA.md for dataset preparation.

We used transfer learning to partly initialize the tunable weights of HierAttn and SOTA models. Please refer to PREMODEL.md for pretrained models download.

Skin lesions classification in dermoscopy dataset

IHISIC20000 Val	# Parameters/M	Top-1 Accuracy/%↑
MobileNetV2	2.23	93.45
MobileViT_s	4.94	94.72
MobileNetV3_Large	4.21	94.77
ShuffleNetV2_1×	2.28	95.23
MnasNet1.0	3.11	95.45
EfficientNet_b0	4.02	95.48
HierAttn_xs(Ours)	1.08	96.15
HierAttn_s(Ours)	2.14	96.70

Skin lesions classification in smartphone dataset

IHPAD3000 Val	# Parameters/M	Top-1 Accuracy/%↑
MobileNetV2	2.23	87.44
ShuffleNetV2_1×	2.28	87.89
MobileViT_s	4.94	88.22
MobileNetV3_Large	4.21	88.78
EfficientNet_b0	4.02	90.22
MnasNet1.0	3.11	90.33
HierAttn_xs(Ours)	1.08	90.11
HierAttn_s(Ours)	2.13	91.22

Citation

If you find it useful in your research, please consider citing our paper as follows:

@ARTICLE{10230242,
  author={Dai, Wei and Liu, Rui and Wu, Tianyi and Wang, Min and Yin, Jianqin and Liu, Jun},
  journal={IEEE Journal of Biomedical and Health Informatics}, 
  title={Deeply Supervised Skin Lesions Diagnosis With Stage and Branch Attention}, 
  year={2024},
  volume={28},
  number={2},
  pages={719-729},
  keywords={Skin;Lesions;Feature extraction;Convolution;Transformers;Training;Computational modeling;Attention;deep supervision;disease classification;skin lesion;vision transformer},
  doi={10.1109/JBHI.2023.3308697}}

Acknowledgment

Many thanks to authors of ml-cvnets for their great framework!