Diabetic retinopathy is a common microvascular complication of diabetes that can lead to vision loss. It is typically identified through retinal lesions such as soft exudates, hard exudates, microaneurysms, and hemorrhages. Since diagnosis involves analyzing retinal images, computational models can support experts in identifying these lesions during clinical examinations. Among them, microaneurysms are the earliest detectable signs but present a particular challenge due to their small size and low contrast. These features often appear near the macula, making this region especially relevant for early diagnosis and treatment planning. In this work, we propose an approach based on the YOLO11 architecture to perform both detection and segmentation of retinal lesions, emphasizing the macular region as a region of interest. We train two deep learning models in a sequential pipeline: the first extracts macula-centered regions using the Indian Diabetic Retinopathy Image Dataset; the second detects and segments lesions within the extracted region of interest using the Dataset for Diabetic Retinopathy. Our results show that the lesion detection model achieved a mean average precision (mAP@50) of 0.4460 on the validation set and 0.2830 on the test set of the Dataset for Diabetic Retinopathy, demonstrating the effectiveness of macula-focused region of interest extraction in improving lesion detection performance.