Adaptive PSO with Orientation Awareness for Robust Object Localization and Bounding Box Refinement

This study talks about how an artificial intelligence (AI) melanoma detection system was made and tested. It uses a mix of convolutional neural networks (CNNs) and a minimum spanning tree (MST) algorithm to give a real-time risk assessment. With a 92.30% accuracy, 91.70% precision, 92.50% recall, and an area under the curve (AUC) score of 0.94, the system demonstrated high performance in classifying melanoma lesions and predicting risk.The proposed method makes melanoma risk assessments more accurate by combining visual factors with environmental risk factors like UV, chemical, and bacterial exposure. Because of this integration, lesions can now be categorised more precisely and fully, which expedites the early diagnostic and therapeutic procedure.This system's greater impact lies in its ability to allow edge-cloud AI applications, which in turn enables reliable, real-time melanoma detection, potentially aiding in early diagnosis and preventive therapy. Future studies could focus on enhancing model generalisability, integrating real-time sensor data to improve the accuracy of environmental risk factor estimates, and exploring sophisticated object detection models like YOLO or Faster R-CNN to improve lesion localisation and classification accuracy.