Digital Melanoma Diagnoses

Malignant melanoma is the fifth most common cancer type in Norway and is also among the cancer types with the highest mortality. The incidence has increased by more than 50% since the year 2000. The high mortality rate is largely due to the late presentation of patients with more advanced tumours, and particularly among older men. Primary cancer diagnosis of melanoma relies entirely on the pathologist’s relatively standardised examination of the skin lesion. Digitalisation allows the use of image analysis to provide more robust quantitative data. The Malignant Melanoma Project’s purpose is to examine whether artificial intelligence can be used for prognostication.

In 2018, 2556 persons were diagnosed with cutaneous melanoma, and more than 300 patients die from this disease annually. Five-year survival is approximately 90% for women and 80% for men for all melanomas independent of stage. Most melanomas arise on the skin, although some occur in the mucosa, the eye and the genital region. The incidence of melanomas in the latter groups has been relatively stable for several years. In this project, only melanomas arising on the skin are described and discussed.

The significant increase in malignant melanoma incidence, particularly in Norway, has been discussed. If the disease is over-diagnosed, unnecessary distress and quality of life are likely consequences for the affected persons.

Primary cancer diagnosis relies entirely on the pathologist’s examination. The pathologist performs a relatively standardised examination of the skin lesion. Digitalisation allows the use of image analysis to provide more robust quantitative data. Applications for in-silico pathology are being developed for several cancer types at ICGI. Given the problems described above in melanoma diagnostics, and the serious impact of diagnosis for the individual patient regarding treatment and further follow-up, it is reasonable to investigate whether digital diagnostics with artificial intelligence can improve precision and reproducibility in diagnostics.

The purpose of the Malignant Melanoma Project is to answer the following questions:

• Is it possible to identify morphological characteristics for the thin (<1 mm) melanomas with a poor prognosis, beyond thickness?
• Can artificial intelligence reduce the number of lesions assigned an uncertain and potentially malignant diagnose (atypical melanocytic lesion)?
• How much morphological heterogeneity exists in each tumour, and can we use this information to provide more certain guidelines for the sampling of tissue sections and microscopical sections?

We will identify all melanomas diagnosed with thickness <1 mm at OUS in the period 1998-2018. All lesions will be sectioned into 20 additional sections, classified morphologically, and scanned together with original sections. We will include information about which patients developed metastases, and lesions from these patients will be compared with those who had no metastases.

We will further identify and scan lesions with the diagnosis “Atypical melanocytic lesion” from OUS in the period 1998-2018. We will include information about the cause of death, and evaluate whether artificial intelligence can be utilised to separate images from patients who developed metastases from those that did not, i.e., to develop a method for improved differential diagnostics.

ll superficial spreading and nodular melanomas from OUS in the period 1998-2018 will be identified, 20 additional sections from each of these will be cut, and both the additional sections and the original sections will be scanned. We will then include information about which patients developed metastases and evaluate whether artificial intelligence can predict the patients’ prognosis better than manual examinations with conventional microscopy.