Skin cancer places a huge burden on the Australian health care system, with an estimated cost for diagnosing and treating early and late-stage melanoma alone exceeding AU$300+ million per annum. Early diagnosis of melanoma is however challenging as melanomas often mimic benign lesions such as melanocytic naevi. Once a lesion is excised, the gold standard for diagnosis is through microscopic examination of hematoxylin and eosin (H&E) stained tissue samples. However, examining H&E images is time-consuming for ambiguous cases, variable, and is not sensitive enough to precisely define cell types with a high degree of accuracy. Since naevi and melanoma share many common histopathological features, diagnosis is challenging even for experienced dermatopathologists. To this end, this project aims to accurately identify cell types and transcriptional cell states in benign and malignant lesions. We hypothesize that discrete cell populations in early melanoma tumours, identified as having invasive potential, will be detectable using spatial profiling tools. To address this, we have performed Visium CytAssist on a large progression series of samples ranging from benign naevi, naevi with low to high grade dysplasia, melanoma in situ, melanoma arising from a pre-existing naevus, as well as thin melanomas. To fully characterise these lesions, we have also performed targeted panel sequencing to investigates somatic mutation burden and copy number aberrations. This study will integrate these molecular data with spatial gene expression to develop a deep learning model for melanoma diagnosis from H&E-stained slides. We envision that these data will permit AI-guided diagnosis to accurately determine the benign/malignant status of melanocytic lesions.