Journal: Biomedical Signal Processing and Control

Article Title: Detection of COVID-19 findings by the local interpretable model-agnostic explanations method of types-based activations extracted from CNNs

doi: 10.1016/j.bspc.2021.103128

Figure Lengend Snippet: Sorting the selected types-based activations among CNN models by applying the LIME method to the CPB data.

Article Snippet: Before the last layer of transfer learning-based CNN models (pre-Softmax), activation sets as many as the number of class types are formed.

Techniques:

Journal: Biomedical Signal Processing and Control

Article Title: Detection of COVID-19 findings by the local interpretable model-agnostic explanations method of types-based activations extracted from CNNs

doi: 10.1016/j.bspc.2021.103128

Figure Lengend Snippet: Sorting the selected types-based activations among CNN models by applying the LIME method to the CovCT-Findings data.

Article Snippet: Before the last layer of transfer learning-based CNN models (pre-Softmax), activation sets as many as the number of class types are formed.

Techniques:

Journal: Sensors (Basel, Switzerland)

Article Title: Classification of HEp-2 Staining Pattern Images Using Adapted Multilayer Perceptron Neural Network-Based Intra-Class Variation of Cell Shape

doi: 10.3390/s23042195

Figure Lengend Snippet: Results were achieved by the participants compared in terms of MCA to the recent methods, ICPR2014 and ICPR2016 contests (the first sixteenth rows of the table) and our adapted (MLP) classifier (the remaining rows of the table in bold text) over Task-1 training dataset.

Article Snippet: [ ] , – , CNN-based Softmax , rotation, cropping & flipping , 95.32 , 91.33.

Techniques: Extraction, Selection, Plasmid Preparation

Journal: Journal of Medical Imaging

Article Title: Patient-specific fine-tuning of convolutional neural networks for follow-up lesion quantification

doi: 10.1117/1.JMI.7.6.064003

Figure Lengend Snippet: Proposed lesion quantification framework, shown with the liver MRI as an example. First a base CNN is trained with a training set consisting of multiple patients. Next, the base CNN is refined in the patient-specific FT step using a previous MRI exam of a patient (the baseline scan). The fine-tuned CNN is used to detect or segment lesions in a follow-up MRI scan of the same patient. The images are cropped to focus of the organ of interest. The cropped image size is 128 × 128 pixels .

Article Snippet: The Softmax probabilities of the base CNN had a mean maximum SD of 0.398 ( ± 0.025 ).

Techniques:

Journal: Journal of Medical Imaging

Article Title: Patient-specific fine-tuning of convolutional neural networks for follow-up lesion quantification

doi: 10.1117/1.JMI.7.6.064003

Figure Lengend Snippet: Median (IQR) of the TPR, FPC, and F1 score of the liver metastases detection for a varying number of iterations of learning for the CNN for FT. The best results are printed in bold.

Article Snippet: The Softmax probabilities of the base CNN had a mean maximum SD of 0.398 ( ± 0.025 ).

Techniques:

Journal: Journal of Medical Imaging

Article Title: Patient-specific fine-tuning of convolutional neural networks for follow-up lesion quantification

doi: 10.1117/1.JMI.7.6.064003

Figure Lengend Snippet: Median (IQR) of the TPR, FPC, and F1 score for a ranging number of slices presented to the CNN for FT. The best results are printed in bold. No significant differences were found between the Base CNN and all options.

Article Snippet: The Softmax probabilities of the base CNN had a mean maximum SD of 0.398 ( ± 0.025 ).

Techniques:

Journal: Journal of Medical Imaging

Article Title: Patient-specific fine-tuning of convolutional neural networks for follow-up lesion quantification

doi: 10.1117/1.JMI.7.6.064003

Figure Lengend Snippet: Median (IQR) of the TPR, the FPC and the F1 score of the liver metastases detection, for weighting the true positives, false negatives, and false positives during the patient-specific FT. The best results are printed in bold.

Article Snippet: The Softmax probabilities of the base CNN had a mean maximum SD of 0.398 ( ± 0.025 ).

Techniques:

Journal: Journal of Medical Imaging

Article Title: Patient-specific fine-tuning of convolutional neural networks for follow-up lesion quantification

doi: 10.1117/1.JMI.7.6.064003

Figure Lengend Snippet: Examples of the detection results on the follow-up scan of the base CNN and the patient-specific CNN for three different patients. White outline = manual annotation, red outline = false positive object, green check = detected metastasis, red cross = missed metastasis.

Article Snippet: The Softmax probabilities of the base CNN had a mean maximum SD of 0.398 ( ± 0.025 ).

Techniques:

Journal: Journal of Medical Imaging

Article Title: Patient-specific fine-tuning of convolutional neural networks for follow-up lesion quantification

doi: 10.1117/1.JMI.7.6.064003

Figure Lengend Snippet: Mean ( ± SD ) of the Dice score and AVD of the WMH segmentation for a varying number of slices for FT. The best results are printed in bold.

Article Snippet: The Softmax probabilities of the base CNN had a mean maximum SD of 0.398 ( ± 0.025 ).

Techniques:

Journal: Journal of Medical Imaging

Article Title: Patient-specific fine-tuning of convolutional neural networks for follow-up lesion quantification

doi: 10.1117/1.JMI.7.6.064003

Figure Lengend Snippet: Mean ( ± SD ) of the Dice score and AVD of the WMH segmentation for weighting the true positives, false negatives, and false positives during the patient-specific FT. The best results are printed in bold.

Article Snippet: The Softmax probabilities of the base CNN had a mean maximum SD of 0.398 ( ± 0.025 ).

Techniques:

Journal: Journal of Medical Imaging

Article Title: Patient-specific fine-tuning of convolutional neural networks for follow-up lesion quantification

doi: 10.1117/1.JMI.7.6.064003

Figure Lengend Snippet: Examples of the follow-up scan with the segmentation results of the base CNN and the patient-specific CNN for three different patients. Green = true positive pixels, red = false negative pixels, and blue = false positive pixels.

Article Snippet: The Softmax probabilities of the base CNN had a mean maximum SD of 0.398 ( ± 0.025 ).

Techniques:

Journal: Journal of Medical Imaging

Article Title: Patient-specific fine-tuning of convolutional neural networks for follow-up lesion quantification

doi: 10.1117/1.JMI.7.6.064003

Figure Lengend Snippet: An example of the uncertainty (SD of Softmax probability) of the base CNN and the patient-specific CNN. A high SD means the CNN is uncertain about its decision.

Article Snippet: The Softmax probabilities of the base CNN had a mean maximum SD of 0.398 ( ± 0.025 ).

Techniques: