Journal: NPJ Digital Medicine

Article Title: Advanced and interpretable corneal staining assessment through fine grained knowledge distillation

doi: 10.1038/s41746-025-01706-y

Figure Lengend Snippet: The first row is the original raw image. The second row is the class activation mapping plus plus (CAM++) of the ResNet50. The third row is the fined-grained lesion detection visualization from the proposed FKD-CSS model. FKD fine-grained knowledge distillation. CSS corneal staining score.

Article Snippet: We trained a U-Net on the SUSTech-SYSU dataset utilizing a pre-trained ResNet50 backbone on ImageNet for automatic corneal segmentation.

Techniques: Activation Assay, Distillation, Staining

Journal: International Journal of Computer Assisted Radiology and Surgery

Article Title: EndoViT: pretraining vision transformers on a large collection of endoscopic images

doi: 10.1007/s11548-024-03091-5

Figure Lengend Snippet: Action triplet recognition full dataset results (mAP)

Article Snippet: For reference purposes, we note that the reported performance of ResNet50 backbone in TeCNO [ ] is 88.56% ± 0.27%.

Techniques:

Journal: International Journal of Computer Assisted Radiology and Surgery

Article Title: EndoViT: pretraining vision transformers on a large collection of endoscopic images

doi: 10.1007/s11548-024-03091-5

Figure Lengend Snippet: Action triplet recognition few-shot results (mAP)

Article Snippet: For reference purposes, we note that the reported performance of ResNet50 backbone in TeCNO [ ] is 88.56% ± 0.27%.

Techniques:

Journal: International Journal of Computer Assisted Radiology and Surgery

Article Title: EndoViT: pretraining vision transformers on a large collection of endoscopic images

doi: 10.1007/s11548-024-03091-5

Figure Lengend Snippet: Surgical phase recognition few-shot results (mean accuracy)

Article Snippet: For reference purposes, we note that the reported performance of ResNet50 backbone in TeCNO [ ] is 88.56% ± 0.27%.

Techniques:

Journal: International Journal of Computer Assisted Radiology and Surgery

Article Title: EndoViT: pretraining vision transformers on a large collection of endoscopic images

doi: 10.1007/s11548-024-03091-5

Figure Lengend Snippet: Surgical phase recognition full dataset results (mean Accuracy)

Article Snippet: For reference purposes, we note that the reported performance of ResNet50 backbone in TeCNO [ ] is 88.56% ± 0.27%.

Techniques:

Journal: Scientific Reports

Article Title: Automated detection of the HER2 gene amplification status in Fluorescence in situ hybridization images for the diagnostics of cancer tissues

doi: 10.1038/s41598-019-44643-z

Figure Lengend Snippet: Illustration of the two-stage deep learning detection system of the HER2 gene amplification stage in FISH images from breast cancer samples. ( A ) The nucleus detector network takes whole FISH images as input and outputs the localization and classification for all detected nuclei. ( B ) The signal detector network subsequently takes each detected nucleus and localizes and classifies individual FISH signals. The output of both networks is post-processed by calculation of the low/high grade ratios and HER2/CEN17 ratios, and an image-wide classification prediction is computed and reported. ( C ) Both detectors are based on RetinaNet which consists of a ResNet50 feature extraction network, a feature pyramid network and two fully convolutional classification and box regression networks for every level of the feature pyramid.

Article Snippet: In this work, we used the implementation of RetinaNet provided by Fizyr with a ResNet50 backbone provided by the Broad Institute.

Techniques: Amplification, Extraction