Explaining Eye Diseases Detected by Machine Learning Using SHAP: A Case Study of Diabetic Retinopathy and Choroidal Nevus

Shakeri, Esmaeil; Crump, Trafford; Weis, Ezekiel; Mohammed, Emad A.; Souza, Roberto; Far, Behrouz H.

doi:10.1007/s42979-023-01859-1

Cited by 3 publications

(5 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The proposed model is different from all these works in that it detects external eye diseases, whereas all the other works shown in the table detect internal eye diseases. Methodologically, all the previous researchers used the datasets (1,3,16,18,23,25) of DR, age-based, and gender-based related to internal eye disease. Here, in this article, Ethiopian datasets (12) of external eye disease images and public fundus images were https://www.indjst.org/ used and evaluated.…”

Section: Cross-validation R2 Metric For the Modelmentioning

confidence: 99%

“…• Maximum of the models focused on only one task for the detection of eye diseases, whereas there was the possibility to make the models perform more than one task. (15,16) • The work carried out for the detection of external eye diseases (17) suggested improvements in terms of accuracy and speed of training the model. The new model is required to detect external eye diseases effectively.…”

Section: Introductionmentioning

confidence: 99%

“…https://www.indjst.org/ Furthermore, deep learning (DL) was utilized for the diagnosis process. (16,19) PPA (parapapillary decay) was related to nearsightedness and glaucoma, so the identification of PPA was a substantial indicator for these diseases as well. Glaucoma screening using an optic plate and optic cup division technique based on super-pixel characterization has been studied as an algorithm for detecting retinal arteriovenous (AV) scratching.…”

Section: Introductionmentioning

confidence: 99%

“…(21,22) Overall, they (18) used Vgg19 deep learning with classification produced an accuracy 99.17% for the model. In the case of (16) achieved accuracy of 97% and 81% of predictive and multiclass classification with Shapely Addictive Explanation (SHAP) for diabetic retinopathy (DR), but they used small datasets of 614 only for training and validation. DR images of 1200 men and women were used (16) for a predictive model with ResNet101 and achieved a result of 96.92%.…”

Section: Introductionmentioning

confidence: 99%

“…In the case of (16) achieved accuracy of 97% and 81% of predictive and multiclass classification with Shapely Addictive Explanation (SHAP) for diabetic retinopathy (DR), but they used small datasets of 614 only for training and validation. DR images of 1200 men and women were used (16) for a predictive model with ResNet101 and achieved a result of 96.92%. Researchers used transfer deep learning with the Vgg19 model.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Multiclass Classification and Identification of the External Eye Diseases using Deep CNN

Rashid,

Abate,

Abdi

2024

IJST

View full text Add to dashboard Cite

Objective: These eye illnesses can be either internal eye diseases or external eye diseases. The purpose of research is to find the right model for better performance to identify the external eye disease. The model is customised with 16- layers CNN using multiclass classification. Method: The Deep CNN techniques are utilized with multiclass classification, and the model is developed using Vgg16 with different dropout rates of 0.25 and 0.50 to improve accuracy and performance. In this work, a deep convolutional neural network model is proposed to classify and identify external eye diseases like conjunctivitis, blepharitis, and cellulitis. Datasets were taken in 80:20 randomly from blepharitis, cellulitis, and Conjunctivitis to test (242) and train (968) the model after pre-processing. Novelty: The model is novel and unique using deep CNN, Vgg16, and multiclass classification because it has never been classified and predicted previously for external eye disease. Additionally, Vgg16 with dropout rates of 0.25 and 0.50 was not tested. The model is penetrated into fully connected (FC) layers with different dropout rates. Findings: The accuracy of 98.48% and 0.976% for deep CNN and multiclass classification consecutively produced satisfactory results. The efficiency of R2 is evaluated with multiple classes of data that resulted in a range of 0.425 - 0.775 with k = 10 folds. Vgg16 attains the highest performance of 71.54% with changed dropout rates. The effects of fundus in the ocular, such as retinopathy and AMD, can be examined in the future with segmented data using CNN for better optimization. On account of biological changes in eye and retinal structure, models might be constructed or studied. Keywords: Multiclass Classification, Identification, Deep CNN, External Eye Disease, Evaluation

show abstract