Quantitative and Qualitative Evaluation of Explainable Deep Learning Methods for Ophthalmic Diagnosis

Singh, Amitojdeep; Balaji, J. Jothi; Rasheed, Mohammed Abdul; Jayakumar, Varadharajan; Raman, Rajiv; Lakshminarayanan, Vasudevan

doi:10.48550/arxiv.2009.12648

Cited by 2 publications

(2 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This is especially important because many recent deep learning based clinical studies rely on saliency maps for interpretability of deep learning models without noting and critically evaluating their inherent limitations. A recent empirical study found that ophthalmologists and optometrists rated GBP highly as an explainability method, despite the limitations we note in this study [33]. Table 2 demonstrates the overall results for each saliency map across all tests.…”

Section: Discussionmentioning

confidence: 93%

Assessing the Trustworthiness of Saliency Maps for Localizing Abnormalities in Medical Imaging

Arun¹,

Gaw²,

Singh³

et al. 2021

Radiology: Artificial Intelligence

158

View full text Add to dashboard Cite

To evaluate the trustworthiness of saliency maps for abnormality localization in medical imaging. Materials and Methods:Using two large publicly available radiology datasets (SIIM-ACR Pneumothorax Segmentation and RSNA Pneumonia Detection), we quantified the performance of eight commonly used saliency map techniques in regards to their 1) localization utility (segmentation and detection), 2) sensitivity to model weight randomization, 3) repeatability, and 4) reproducibility. We compared their performances versus baseline methods and localization network architectures, using area under the precision-recall curve (AUPRC) and structural similarity index (SSIM) as metrics.Results: All eight saliency map techniques fail at least one of the criteria and were inferior in performance compared to localization networks. For pneumothorax segmentation, the AUPRC ranged from 0.024-0.224, while a U-Net achieved a significantly superior AUPRC of 0.404 (p<0.005). For pneumonia detection, the AUPRC ranged from 0.160-0.519, while a RetinaNet achieved a significantly superior AUPRC of 0.596 (p<0.005). Five and two saliency methods (out of eight) failed the model randomization test on the segmentation and detection datasets, respectively, suggesting that these methods are not sensitive to changes in model parameters. The repeatability and reproducibility of the majority of the saliency methods were worse than localization networks for both the segmentation and detection datasets. Conclusion:We suggest that the use of saliency maps in the high-risk domain of medical imaging warrants additional scrutiny and recommend that detection or segmentation models be used if localization is the desired output of the network.

show abstract

Section: Discussionmentioning

confidence: 93%

Assessing the Trustworthiness of Saliency Maps for Localizing Abnormalities in Medical Imaging

Arun¹,

Gaw²,

Singh³

et al. 2021

Radiology: Artificial Intelligence

158

View full text Add to dashboard Cite

show abstract

“…place, arrange pictures given the presence or nonappearance of DME then pass positive cases into the second stage to mark them in light of seriousness. The model accomplished 96.12%, 96.32%, 95.84%, and an F−1 score of 96% for Accuracy, Sensitivity, Specificity, and F−1 score, separately[22]. Wang et al proposed a clever calculation named SBGFRLS-OCT calculation for the division and location of DME in an OCT picture.…”

mentioning

confidence: 99%

A Deep Learning Approach for Detecting Diabetic Macular Edema through Analyzing Retinal Images

Mishra¹

2022

MSEA

View full text Add to dashboard Cite

Clinical imaging developed quickly to assume an imperative part in the conclusion and treatment of an illness. Robotized examination of clinical picture examination has expanded successfully using profound learning procedures to get much speedier groupings once prepared and learn significant highlights for explicit assignments, demonstrated to be assessable in clinical practice and an important device to help dynamic in the clinical field. Inside Opthalmology, Optical Coherence Tomography (OCT) is a volumetric imaging methodology that purposes the conclusion, observing, and estimating reaction to treatment in the eyes. Early discovery of eyes sicknesses including Diabetic Macular Edema (DME) is crucial interaction to keep away from confusion like visual impairment. This work utilized a profound convolutional brain organization (CNN) based technique for the DME order tasks. To exhibit the effect of convolutional, five models with various Convolutional layers were assembled then the best one chose given assessment measurements. The exactness of the model improved while expanding the quantity of Convolutional Layers and accomplished 82% by 5-Convolutional Layer, Precision and Recall of the CNN model per DME class were 87%% and 74%, individually. These outcomes featured the capability of profound learning in helping dynamics in patients with DME.

show abstract

Quantitative and Qualitative Evaluation of Explainable Deep Learning Methods for Ophthalmic Diagnosis

Cited by 2 publications

References 0 publications

Assessing the Trustworthiness of Saliency Maps for Localizing Abnormalities in Medical Imaging

Assessing the Trustworthiness of Saliency Maps for Localizing Abnormalities in Medical Imaging

A Deep Learning Approach for Detecting Diabetic Macular Edema through Analyzing Retinal Images

Contact Info

Product

Resources

About