Development and validation of prognostic factors and a deep learning algorithm in uveal melanoma

The uveal tract comprises three anatomical structures: the iris, the ciliary body, and the choroid. Uveal melanoma (UM) is the most common primary malignant intraocular tumor in adults and has a high propensity for metastasis. Unfortunately, once radiologically detectable metastases develop, the median patient survival is only about 1.5 years. No major improvements in patient survival have been achieved during the last several decades, making it essential to identify factors that can help improve outcomes. Detecting lesions at the earliest possible stage-before they grow large-and accurately distinguishing the few truly malignant lesions from the larger number of benign counterparts is likely an effective way to enhance survival.

Recognizing the factors that influence disease progression is critical for improving prognosis. Without this understanding, meaningful progress remains limited. Therefore, the main focus of this dissertation is to investigate clinical factors that help us better understand the disease, as well as to identify technological tools that can aid in earlier diagnosis. By gaining deeper insight, we aim to improve patient outcomes and make a meaningful impact on the course of the disease.

In paper I, we compared the long-term prognosis for patients with iris melanomas with small choroidal melanomas. It has previously been described that iris melanomas have a more favorable prognosis compared to choroidal melanomas. However, iris melanomas are typically relatively small at diagnosis, which could contribute to their favorable prognosis. Tumor diameter, which is one of the strongest predictors of uveal melanoma prognosis, has not been taken into full account in previous studies. We have therefore compared the two tumor types and adjusted for tumor size. Our findings did not show any survival differences between iris and choroidal melanomas.

In paper II, we investigated the association between vasculogenic mimicry (VM), presenting symptoms, patient outcome and the area density of periodic acid-Schiff positive histological patterns (PAS density). PAS stains a range of tissues including VM, which is a fluid conducting extracellular matrix pattern that has in many cancers been linked to worse prognosis. One of our previous studies observed that patients with uveal melanoma having a shadow in the visual field as a symptom were more likely to have retinal detachment and worse prognosis. In this study, we demonstrated that the likely reason a shadow in the visual field is associated with a worse prognosis is that these tumors are more prone to have higher PAS density and display VM.

In paper III, we wanted to investigate the relationship between obesity and metabolic factors with uveal melanoma prognosis. High body mass index (BMI) is generally believed to be linked to a less favorable prognostic factor. However, our results showed that obesity and high serum levels of leptin were associated with a favorable prognosis, described as an obesity paradox.

In paper IV, we developed a deep learning algorithm to distinguish small choroidal melanomas from nevi. Early detection in melanomas is crucial, as even small melanomas can metastasize and become life threatening but detecting them at an early stage is often challenging. Additionally, the task of distinguishing these lesions often falls on non-experts, making decision supporting tools even more valuable. Our algorithm not only matched but, in some cases, outperformed human ophthalmologists, making it a potential tool for clinicians in determining tumor monitoring frequencies and deciding when to refer for further evaluation or treatment.

List of scientific papers

I. Sabazade S, Herrspiegel C, Gill V, Stålhammar G. No differences in the long-term prognosis of iris and choroidal melanomas when adjusting for tumor thickness and diameter. BMC Cancer. 2021 Nov 24;21(1):1270. Erratum in: BMC Cancer. 2022 Jan 4;22(1):32. doi: 10.1186/s12885-021-09167- 8.
https://doi.org/10.1186/s12885-021-09002-0

II. Sabazade S, Gill V, Herrspiegel C, Stålhammar G. Vasculogenic mimicry correlates to presenting symptoms and mortality in uveal melanoma. J Cancer Res Clin Oncol. 2022 Mar;148(3):587-597.
https://doi.org/10.1007/s00432-021-03851-9

III. Sabazade S, Opalko A, Herrspiegel C, Gill VT, Plastino F, André H, Stålhammar G. Obesity paradox in uveal melanoma: high body mass index is associated with low metastatic risk. Br J Ophthalmol. 2024 Mar 20;108(4):578-587.
https://doi.org/10.1136/bjo-2022-322877

IV. Sabazade S, Lumia Michalski MA, Bartoszek J, Fili M, Holmström M, Stålhammar G. Development and Validation of a Deep Learning Algorithm for Differentiation of Choroidal Nevi from Small Melanoma in Fundus Photographs. Ophthalmol Sci. 2024 Aug 30;5(1):100613.
https://doi.org/10.1016/j.xops.2024.100613