<scp>MEDCnet</scp> : A Memory Efficient Approach for Processing High‐Resolution Fundus Images for Diabetic Retinopathy Classification Using <scp>CNN</scp>
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
ABSTRACT Modern medical imaging equipment can capture very high‐resolution images with detailed features. These high‐resolution images have been used in several domains. Diabetic retinopathy (DR) is a medical condition where increased blood sugar levels of diabetic patients affect the retinal vessels of the eye. The usage of high‐resolution fundus images in DR classification is quite limited due to Graphics processing unit (GPU) memory constraints. The GPU memory problem becomes even worse with the increased complexity of the current state‐of‐the‐art deep learning models. In this paper, we propose a memory‐efficient divide‐and‐conquer‐based approach for training deep learning models that can identify both high‐level and detailed low‐level features from high‐resolution images within given GPU memory constraints. The proposed approach initially uses the traditional transfer learning technique to train the deep learning model with reduced‐sized images. This trained model is used to extract detailed low‐level features from fixed‐size patches of higher‐resolution fundus images. These detailed features are then utilized for classification based on standard machine learning algorithms. We have evaluated our proposed approach using the DDR and APTOS datasets. The results of our approach are compared with different approaches, and our model achieves a maximum classification accuracy of 95.92% and 97.39% on the DDR and APTOS datasets, respectively. In general, the proposed approach can be used to get better accuracy by using detailed features from high‐resolution images within GPU memory constraints.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it