Steven Lawrence Fernandes

Background/Objectives: Disorganization of retinal inner layers (DRIL) is an important and supportive biomarker in optical coherence tomography (OCT) imaging for diagnosing the extent of diabetic macular edema (DME) in patients and anticipating visual outcomes. But the manual DRIL identification is subject to interobserver bias and requires a lot of time and effort from the experts. This research presents a novel, computerized, and clinically guided approach for the classification of DRIL that leverages the central 1 mm foveal region extracted through the annotations provided by the expert ophthalmologists and investigates the effectiveness of a transformer and Masked Auto Encoder (MAE) based foundation model (RETFound) as the primary approach. Methods: We fine-tuned and validated the RETFound model, utilizing accurate foveal center coordinates provided by the experienced ophthalmologists. Our approach emphasizes the macular region that is significant diagnostically, where DME biomarkers manifest more predominantly. To guarantee robust evaluation, the dataset was divided into 85% training and 15% held-out test sets. We performed 5-fold cross-validation exclusively on the training dataset with baseline, conservative, and moderate fine-tuning strategies, and the final model was evaluated on the independent, unseen test set. Convolutional neural network (CNN)-based transfer learning (TL) models (MobileNetV2, EfficientNetB0, InceptionV3, DenseNet121, and DenseNet169) were also assessed for comparative evaluation. Results: The RETFound model yielded the best outcomes under the conservative fine-tuning strategy, achieving a mean test accuracy (AC) of 0.9339 ± 0.0036 and an area under the curve (AUC) of 0.9660 ± 0.0028 on the independent held-out test set across the five fold-trained models. The moderate and baseline evaluations achieved comparatively lower outcomes, highlighting the effectiveness of the conservative approach. The RETFound model consistently outperformed CNN models, exhibiting stability and superior generalization for DRIL classification. We performed statistical validation using the Wilcoxon signed-rank test and 95% confidence intervals to confirm the robustness of the proposed method, and an ablation analysis showed that the fovea-centered region of interest (ROI) guidance consistently improved results when compared with whole OCT analysis. Conclusions: This research demonstrates that the deep-learning (DL) methods assisted by expert clinical knowledge with an anatomically aligned ROI could provide remarkable results in DRIL detection applications. This work attempts to establish an anatomically relevant framework for computerized DRIL identification that focuses on the highly crucial macular region, possibly helping in faster intervention and improved diagnosis in the management of DME.

Background/Objectives: Hyperreflective foci (HRF) are supportive optical coherence tomography (OCT) imaging biomarkers that have been examined for their association with disease progression and severity in various retinal disorders. The accurate identification and segmentation of these tiny structures of lipid extravasation remain complicated because of their small size, class imbalance, similarity in the reflectivity patterns with the surrounding structures and imaging artifacts. While U-Net-based models have promised exceptional results for medical image segmentation, optimal architectural settings and suitable preprocessing methods for HRF detection remain unclear. Methods: This research assessed optimal settings for U-Net-based models for HRF segmentation by evaluating standard U-Net and attention U-Net under different preprocessing regimes. Attention U-Net employed Z-score normalization and contrast-limited adaptive histogram equalization (CLAHE) enhancement with soft dice loss. The standard U-Net was trained on OCT images with CLAHE using focal Tversky loss. A total of 435 fovea-centered OCT B scans with the corresponding, consensus-annotated HRF masks were utilized for this research. Results: The standard U-Net outperformed attention U-Net with a dice score of 0.5207, an AUC of 0.8411, and a recall of 0.6439 on raw OCT images. The attention U-Net with preprocessing (dice: 0.5033, AUC: 0.6987, recall: 0.5391) demonstrated satisfactory performance. The results showed that the U-Net model with CLAHE and focal Tversky loss improved recall by 19.4% relative to the attention U-Net, and this corresponds roughly to a 23% relative decline in false negatives. This indicates increased sensitivity in identifying HRF regions. Conclusions: The best-performing configuration using U-Net-based architectures for segmentation of HRFs combines the standard U-Net model with CLAHE and focal Tversky loss for handling class imbalance. This approach yields relatively higher sensitivity, indicating that the standard U-Net model delivers a simple and robust framework for automated HRF segmentation on the evaluated dataset, promising further validation in broader clinical datasets.

Background/Objectives: Disorganization of the retinal inner layers (DRIL) is an important biomarker of diabetic macular edema (DME) that has a very strong association with visual acuity (VA) in patients. But the unavailability of annotated training data from experts severely limits the adaptability of models pretrained on real-world images owing to significant variations in the domain, posing two primary challenges for the design of efficient computerized DRIL detection methods. Methods: In an attempt to address these challenges, we propose a novel, self-supervision-based learning framework that employs a huge unlabeled optical coherence tomography (OCT) dataset to learn and detect clinically applicable interpretations before fine-tuning with a small proprietary dataset of annotated OCT images. In this research, we introduce a spatial Bootstrap Your Own Latent (BYOL) with a hybrid spatial aware loss function aimed to capture anatomical representations from unlabeled OCT dataset of 108,309 images that cover various retinal abnormalities, and then adapt the learned interpretations for DRIL classification employing 823 annotated OCT images. Results: With an accuracy of 99.39%, the proposed two-stage approach substantially exceeds the direct transfer learning models pretrained on ImageNet. Conclusions: The findings demonstrate the efficacy of domain-specific self-supervised learning for rare retinal pathological detection tasks with limited annotated data.

Deep learning architectures like ResNet and Inception have produced accurate predictions for classifying benign and malignant tumors in the healthcare domain. This enables healthcare institutions to make data-driven decisions and potentially enable early detection of malignancy by employing computer-vision-based deep learning algorithms. These CNN algorithms, in addition to requiring huge amounts of data, can identify higher- and lower-level features that are significant while classifying tumors into benign or malignant. However, the existing literature is limited in terms of the explainability of the resultant classification, and identifying the exact features that are of importance, which is essential in the decision-making process for healthcare practitioners. Thus, the motivation of this work is to implement a custom classifier on the ovarian tumor dataset, which exhibits high classification performance and subsequently interpret the classification results qualitatively, using various Explainable AI methods, to identify which pixels or regions of interest are given highest importance by the model for classification. The dataset comprises CT scanned images of ovarian tumors taken from to the axial, saggital and coronal planes. State-of-the-art architectures, including a modified ResNet50 derived from the standard pre-trained ResNet50, are implemented in the paper. When compared to the existing state-of-the-art techniques, the proposed modified ResNet50 exhibited a classification accuracy of 97.5 % on the test dataset without increasing the the complexity of the architecture. The results then were carried for interpretation using several explainable AI techniques. The results show that the shape and localized nature of the tumors play important roles for qualitatively determining the ability of the tumor to metastasize and thereafter to be classified as benign or malignant.

Deep learning architectures like ResNet and Inception have produced accurate predictions for classifying benign and malignant tumors in the healthcare domain. This enables healthcare institutions to make data-driven decisions and potentially enable early detection of malignancy by employing computer-vision-based deep learning algorithms. These Convolutional Neural Network algorithms, in addition to requiring huge amounts of data, can identify higher and lower-level features that are significant while classifying tumors into benign or malignant. However, these algorithms have limitations concerning explainability and identifying the exact features that are of importance and contributing to the final classification of tumors as benign or malignant. In this paper, we implement several explainable AI techniques, namely, LIME, Saliency Map, Occlusion Analysis, Grad-CAM, SHAP, and Smooth Grad to interpret the results of a custom ResNet60 classifier in the classification of ovarian tumors as benign or malignant. The ResNet60 model attained an accuracy of 97.50% for the test dataset.

Ovarian cancer is one of the leading causes of death worldwide among the female population. Early diagnosis is crucial for patient treatment. In this work, our main objective is to accurately detect and classify ovarian cancer. To achieve this, two datasets are considered: CT scan images of patients with cancer and those without, and biomarker (clinical parameters) data from all patients. We propose an ensemble deep neural network model and an ensemble machine learning model for the automatic binary classification of ovarian CT scan images and biomarker data. The proposed model incorporates four convolutional neural network models: VGG16, ResNet 152, Inception V3, and DenseNet 101, with transformers applied for feature extraction. These extracted features are fed into our proposed ensemble multi-layer perceptron model for classification. Preprocessing and CNN tuning techniques such as hyperparameter optimization, data augmentation, and fine-tuning are utilized during model training. Our ensemble model outperforms single classifiers and machine learning algorithms, achieving a mean accuracy of 98.96%, a precision of 97.44%, and an F1-score of 98.7%. We compared these results with those obtained using features extracted by the UNet model, followed by classification with our ensemble model. The transformer demonstrated superior performance in feature extraction over the UNet, with a mean Dice score and mean Jaccard score of 0.98 and 0.97, respectively, and standard deviations of 0.04 and 0.06 for benign tumors and 0.99 and 0.98 with standard deviations of 0.01 for malignant tumors. For the biomarker data, the combination of five machine learning models—KNN, logistic regression, SVM, decision tree, and random forest—resulted in an improved accuracy of 92.8% compared to single classifiers.