<p>This study proposes a novel methodology for detecting chemical-induced retinal damage through fundus imaging by leveraging the correlation between retinal abnormalities and chemical exposures. Fundus imaging, a noninvasive and cost-efficient diagnostic modality, is extensively used for detecting ophthalmic and systemic conditions. Pathological manifestations commonly include vascular abnormalities in blood vessels, optic disk deformities, and lesions such as hemorrhages and exudates. Accurate identification and segmentation of these features are crucial for clinically evaluating ocular anomalies. Many pathological conditions share similar manifestations associated with chemical exposure, common to well-known retinal diseases. This paper introduces a multi-scale residual attention U-Net model for retinal feature segmentation and comprehensively maps retinal abnormalities to chemical exposure. To capture broader contextual information, dilated convolutions with layer-specific dilation rates are incorporated into parallel paths within each layer of the U-Net architecture. Furthermore, the convolutional block attention module (CBAM) is integrated into each path to dynamically emphasize spatial and channel-wise features. We also introduce a weighted auxiliary loss function that utilizes output from each decoder block, forcing the model to learn features at a lower-level feature space and enhancing overall system performance. The proposed method is evaluated on three publicly available datasets: DRIVE and STARE for blood vessel segmentation and IDRiD for segmenting exudates, optic disks, and hemorrhages. Performance metrics, including specificity, sensitivity, accuracy, and AUC score, are employed to assess the model. The proposed method can be adapted for automated retinal screening systems for early detection of chemical exposure for public health surveillance and occupational health monitoring.</p>