Hanung Adi Nugroho

Pairwise aggregation methods, such as the Copeland rule, are widely used in collaborative filtering recommender systems due to their robustness and training-free characteristics. However, because these methods rely solely on ordinal pairwise win counts, they frequently produce identical aggregation scores, thereby limiting ranking differentiation. Approaches based purely on cardinal preference information can improve score differentiation but often exhibit unstable recommendation behaviour under data sparsity. This study proposes Ordinal-Cardinal Copeland Aggregation (OCCA), a training-free aggregation framework that integrates ordinal and cardinal preference information within a unified pairwise ranking formulation. The framework integrates pairwise win counts and mean rating differences to improve ranking differentiation while preserving the robustness of majority-based aggregation. Experiments on the MovieLens 100K and MovieLens 1M datasets show that OCCA consistently improves ranking differentiation, achieving Unique Score Ratio (USR) improvements exceeding 56% and 100%, respectively, relative to standard Copeland aggregation. Statistical testing further indicates that these gains are achieved without statistically significant degradation in recommendation accuracy. In addition, OCCA maintains computational efficiency comparable to standard Copeland aggregation.

Glaucoma is an eye disease that can cause permanent blindness without any initial symptoms. Thus, early diagnosis is crucial. The diagnosis is performed by analyzing retinal fundus images, which involves segmentation of the optic disc (OD) and the optic cup (OC), followed by measurement of the cup-to-disc ratio (CDR). Manual analysis by experts is expensive and time-consuming. Many studies used deep learning to automate the process, but most focused only on segmentation or direct classification. Therefore, we developed a system to automatically analyze retinal fundus images. Our main contribution is the utilization of advanced deep learning architectures and backbones for the segmentation, along with a detailed assessment of various CDR measurement methods. Our proposed system consists of pre-processing, segmentation, post-processing, and CDR measurement. During pre-processing, we extracted the region of interest (RoI) using the circular Hough transform. For segmentation, we utilized UNet++ and DeepLabv3+ architectures, with ResNet50, MobileNetv2, and EfficientNetB3 as backbone models. During post-processing, we refined the segmentation results and detected potential issues. Finally, we evaluated various methods to measure CDR. Our proposed system achieved good performance and outperformed other studies in CDR measurement. We used DRISHTI-GS as our primary dataset and RIM-ONE DL as an additional dataset. On the DRISHTI-GS dataset, our RoI extraction achieved a 98% success rate, segmentation using DeepLabv3+ with EfficientNetB3 achieved a Dice score of 92.5% for OD and 89.7% for OC, while vertical CDR measurement achieved a mean absolute error of 0.0385. These results suggest that our proposed system can help with glaucoma diagnosis and provide an efficient, cost-effective approach compared to manual diagnosis.

Diabetic retinopathy (DR) is a leading cause of preventable blindness worldwide, and deep learning systems have achieved competitive performance for DR screening. However, the lack of transparent visual explanations remains a barrier to clinical adoption. Class Activation Map (CAM) methods are widely used to visualize model attention on fundus images, yet their behavior, evaluation, and clinical validity in DR remain fragmented across the literature. We conducted a PRISMA-ScR–guided scoping review of CAM-based visual explanation methods applied to DR classification and lesion-level analysis on color fundus photographs. We searched PubMed, IEEE Xplore, Scopus, Web of Science, and arXiv from 2016 to 2024, using combinations of keywords related to “diabetic retinopathy”, “class activation map”, and “explainable AI”. Studies were included if they applied CAM-family techniques to DR grading or lesion detection on retinal fundus images. The results are a total of 28 studies that met our criteria, covering 14 public and private datasets, predominantly EyePACS, Messidor, APTOS, IDRiD, and FGADR. We summarize the use of Grad-CAM, Grad-CAM++, Score-CAM, Ablation-CAM, LayerCAM, and several DR-specific CAM variants across common backbones (e.g., ResNet, EfficientNet, DenseNet) and categorize evaluation metrics into pixel- and lesion-level localization (IoU, Dice, DAUC/IAUC, ADD, DC/IC, ADCC) and causal faithfulness. We identify recurrent failure modes such as vessel-focused heatmaps, optic disc glare, and poor sensitivity to low-contrast microaneurysms. This review provides an evidence map of CAM-based explanations in DR, highlights gaps between visual plausibility and causal faithfulness, and proposes a compact “what works when” guideline relating DR task, dataset, backbone, and CAM choice. We present a clinical validity lens for lesion-level fidelity and referable DR triage, along with practical recommendations for standardized evaluation and clinically grounded CAM development in DR.

JAIT is a scientific open access journal which focuses on empirical research results and critical analysis of technology development, use, management and impacts in information technology.

This paper proposes a high-capacity reversible watermarking method using adaptive bit-level expansion and pixel-class-guided shifting. Pixels are classified into expandable (P0) and non-expandable (P1) according to their 2-bit LSB patterns, and a lightweight reversible transformation converts P1 into P0 with minimal distortion. A shifting map enables exact recovery and avoids overflow/underflow. Secret data are embedded through a 2-bit LSB expansion rule that ensures full reversibility. Experiments on common and medical images demonstrate a consistent embedding capacity of 1.0 bpp, achieving PSNR values above 46 dB and SSIM above 0.97. In addition, the scheme exhibits low computational overhead (380 kbps) while preserving the original histogram distribution. These results demonstrate that the proposed scheme provides an effective balance between embedding capacity, visual quality, and computational efficiency for secure medical imaging and authenticated reversible data embedding.

This study develops meta-learners for few-shot weakly-supervised segmentation (FWS) to address the challenge of optic disc (OD) and optic cup (OC) segmentation for glaucoma diagnosis with limited labeled fundus images. We significantly improve existing meta-learners by introducing Omni meta-training which balances data usage and diversifies the number of shots. We also develop their efficient versions that reduce computational costs. In addition, we develop sparsification techniques that generate more customizable and representative scribbles and other sparse labels. After evaluating multiple datasets, we find that Omni and efficient versions outperform the original versions, with the best meta-learner being Efficient Omni ProtoSeg (EO-ProtoSeg). It achieves intersection over union (IoU) scores of 88.15% for OD and 71.17% for OC on the REFUGE dataset using just one sparsely labeled image, outperforming few-shot and semi-supervised methods which require more labeled images. Its best performance reaches 86.80% for OD and 71.78%for OC on DRISHTIGS, 88.21% for OD and 73.70% for OC on REFUGE, 80.39% for OD and 52.65% for OC on REFUGE. EO-ProtoSeg is comparable to unsupervised domain adaptation methods yet much lighter with less than two million parameters and does not require any retraining.