Table 1.
Comparative Summary of Related Work.
| Ref | Paper/Year | Methodology Used | Encryption/Security Technique | Attack Detection Model | Limitation |
|---|---|---|---|---|---|
| [1] | Lin et al., Chin. Phys. B, 2026 | CIEA-4DALHS color encryption | 4D augmented Lü hyperchaotic system, spiral scrambling, bit-plane substitution | None | No attack detection; not validated on medical images |
| [2] | Zou et al., Signal Processing, 2026 | Attack-resilient Unet watermarking | Adaptive weighting + RISFT module | None (watermark robustness only) | Watermarking only, no encryption/confidentiality, no attack classification |
| [3] | Zhou et al., Int. J. Theor. Phys., 2024 | Quantum encryption + watermarking | QBM + QDCT + 3D Henon hyperchaos | None | Quantum primitives impractical for real-time clinical use; no attack classifier |
| [4] | Zou et al., Neurocomputing, 2025 | HRFMS watermarking model | Multi-scale convolution, attention mask | None | Watermarking-only scope; no encryption or attack detection |
| [5] | Zhang et al., Digital Signal Processing, 2026 | 4D memristor chaotic encryption | Sprott-C-based memristor system, CNN key generation | None | Encryption-only; no attack-detection/classification stage |
| [6] | Hu et al., Expert Systems with Applications, 2026 | CGI + QMPDFrAT encryption-authentication | Dual chaotic systems, quaternion transform | Authentication only (no classification) | Computationally heavy; no attack-type classification |
| [7] | Gong et al., J. Modern Optics, 2026 | Pixel adaptive diffusion encryption-authentication | Cipherbook-based diffusion, SVDGI | Authentication only | No attack-type classification; ghost-imaging overhead |
| [8] | Veerasekharreddy et al., Neurocomputing, 2025 (Base Paper) | Hyperchaotic Fibonacci polynomial CNN | Hyperchaotic encryption | CNN-based attack classifier | No bio-inspired hyperparameter optimization |
| [9] | Mahalakshmi & Nagarajan, Frontiers in AI, 2026 | DRL-based reversible encryption | DQN-driven adaptive chaotic encryption | None | No attack classification; RL policy stability concerns |
| [10] | Subathra & Thanikaiselvan, Scientific Reports, 2025 | 5D hyperchaotic + U-Net segmentation | 5D hyperchaos, dynamic DNA encoding, zig-zag scrambling | None | No attack detection; segmentation adds processing overhead |
Figure 1.
Overall architecture of the proposed AegisSentinel system.
Figure 2.
Overall architecture of the proposed HFQE-S image encryption framework.
Table 2.
Clean and Attacked Image Variations.
| Medical Image | 0: Clean | 1: Gaussian | 2: Salt&Pepper | 3: Intensity | 4: Occlusion | 5: FGSM |
|---|---|---|---|---|---|---|
 |  |  |  |  |  |  |
Figure 3.
Data acquisition and preprocessing pipeline.
Table 3.
Encryption Performance Across Images.
| Image | Block Size | Rounds | PSNR (dB) | Entropy (bits) |
|---|---|---|---|---|
 | 4×4 | 4 | 6.52 | 7.9973 |
 | 2×2 | 4 | 9.15 | 7.9972 |
 | 3×3 | 3 | 8.00 | 7.9971 |
 | 3×3 | 3 | 6.68 | 7.9974 |
 | 4×4 | 2 | 5.38 | 7.9969 |
Table 4.
HFQE-S Encryption Module Performance Metrics.
| Metric | Achieved Value | Ideal |
|---|---|---|
| Entropy (bits/pixel) | 7.9973 | 8.0000 |
| PSNR (Original vs. Encrypted) | Very Low (noisy) | Low |
| PSNR (Decrypted vs. Original) | Infinite (lossless) | Infinite |
| Encryption Time | 2.87 ms | < 5 ms |
| Decryption Time | 2.35 ms | < 5 ms |
Table 5.
Attack Detection Performance Comparison.
| Configuration | Acc. (%) | Prec. (%) | Rec. (%) | F1 (%) |
|---|---|---|---|---|
| HAPCNN only (No optimizer) | 83.94 | 84.13 | 83.92 | 83.80 |
| HAPCNN + PSO | 89.94 | 89.13 | 89.92 | 89.80 |
| HAPCNN + WOA | 92.51 | 92.55 | 92.38 | 92.71 |
| HAPCNN + COA (Proposed) | 96.52 | 96.60 | 96.48 | 96.49 |
Figure 4.
Confusion matrix of the HAPCNN + COA model on the corrected dataset (1,293 test images, 6 classes).
Table 6.
Per-Class Metrics (Macro-Averaged).
| Class | TP | Row | Col | Precision | Recall | F1 |
|---|---|---|---|---|---|---|
| Clean | 224 | 224 | 235 | 0.9532 | 1.0000 | 0.9760 |
| Gaussian | 208 | 223 | 225 | 0.9244 | 0.9327 | 0.9281 |
| Salt & Pepper | 177 | 182 | 177 | 1.0000 | 0.9725 | 0.9861 |
| Intensity | 225 | 230 | 225 | 1.0000 | 0.9783 | 0.9890 |
| Occlusion | 223 | 223 | 223 | 1.0000 | 1.0000 | 1.0000 |
| FGSM | 191 | 211 | 208 | 0.9183 | 0.9052 | 0.9117 |
| Macro Average | - | - | - | 0.9660 | 0.9648 | 0.9652 |
Table 7.
Comparative analysis of proposed work.
| Encryption Technique | Entropy (bits/pixel) | Encrypt Time | Attack Detection |
|---|---|---|---|
| 4D Memristor Chaotic + CNN Key Gen [17] | 7.9986 | 0.23 s | None |
| CGI + QMPDFrAT + Dual Chaotic [18] | 7.9984 | Not reported | Authentication only |
| Pixel Adaptive Diffusion + SVDGI [19] | ~7.998 | Not reported | Source verification only |
| Hyperchaotic Fibonacci Q-Matrix + HAPCNN-COA [20] | 7.9971 | 2.58 ms | COA-tuned HAPCNN (binary tamper check) |
| HFQE-S (SHA-256 + Fibonacci Q-Matrix) | 7.9973 | 2.87 ms | COA-HAPCNN (6-class) [96.52%] |