Academic Journal
Algorithmic Compression Learning: A Hybrid Complexity–Deep Representation Framework for Medical Image Analysis
| Τίτλος: | Algorithmic Compression Learning: A Hybrid Complexity–Deep Representation Framework for Medical Image Analysis |
|---|---|
| Συγγραφείς: | Mjahed, Soukaina, Mjahed, Ouail |
| Πηγή: | INTERNATIONAL JOURNAL OF COMPUTERS COMMUNICATIONS & CONTROL; Vol. 21 No. 5 (2026): International Journal of Computers Communications & Control (October) ; 1841-9844 ; 1841-9836 ; 10.15837/ijccc.2026.5 |
| Στοιχεία εκδότη: | Agora University Press |
| Έτος έκδοσης: | 2026 |
| Συλλογή: | Agora University Editing House: Journals |
| Θεματικοί όροι: | Algorithmic Complexity, Compression-Based Features, Hybrid Deep Learning, Structural Pattern Analysis |
| Περιγραφή: | Understanding and quantifying structural complexity in visual data remains a fundamental challenge in pattern recognition. This work investigates the hypothesis that compression-based measures can provide a complementary structural representation that enhances deep learning models. To this end, we introduce a hybrid framework termed Algorithmic Compression Learning with Deep Learning (ACL–DL), which integrates multi-scale compression-based descriptors with deep neural representations. The proposed ACL component characterizes structural irregularity and stability across spatial scales using compression-derived measures, interpreted as empirical proxies of structural organization. These descriptors are fused with semantic embeddings extracted from fine-tuned deep networks, enabling joint modeling of intrinsic structural patterns and highlevel visual features. Experiments conducted on three heterogeneous medical imaging datasets (BUSI ultrasound, CBIS-DDSM mammography, and BreakHis histopathology) demonstrate that compression-based descriptors provide discriminative signals that complement deep representations. The best accuracy rates achieved by the ACL–DL hybrid framework, reported as mean + standard deviation, reached 95.63%, 98.72%, and 99.16% across the three datasets, consistently improving over standalone ACL and deep learning baselines with statistically significant gains (p < 0.05). Beyond predictive performance, the results highlight the potential of compression-based structural descriptors as model-agnostic components that can be effectively integrated into modern representation learning pipelines. |
| Τύπος εγγράφου: | article in journal/newspaper |
| Περιγραφή αρχείου: | application/pdf |
| Γλώσσα: | English |
| Relation: | https://univagora.ro/jour/index.php/ijccc/article/view/7478/2616; https://univagora.ro/jour/index.php/ijccc/article/view/7478 |
| DOI: | 10.15837/ijccc.2026.5.7478 |
| Διαθεσιμότητα: | https://univagora.ro/jour/index.php/ijccc/article/view/7478 https://doi.org/10.15837/ijccc.2026.5.7478 |
| Rights: | Copyright (c) 2026 Soukaina MJAHED, Ouail MJAHED ; http://creativecommons.org/licenses/by-nc/4.0 |
| Αριθμός Καταχώρησης: | edsbas.5AC59BA0 |
| Βάση Δεδομένων: | BASE |
καταχωρήστε σχόλιο πρώτοι!