Advancing non-invasive melanoma diagnostics with deep learning and multispectral photoacoustic imaging

Comprehensive Summary

This article discusses the development of non-invasive cutaneous melanoma diagnosis using multispectral photoacoustic imaging and deep learning. Photoacoustic imaging utilizes a laser and high-frequency ultrasound to indirectly characterize the structure and absorption of tissue, to assess the Breslow’s depth of melanoma skin tumor noninvasively. Merdasa et al. applied a computation framework that combined K-means clustering, a one-dimensional convolutional neural network, and an active contour algorithm to photoacoustic data to assess its ability to assess the tumors 3-dimensionally. The model was able to successfully predict Breslow’s depth of the tumors, often overestimating for safety, and define tumor borders without clinician input. It was able to create reconstructions of the melanomas in 3 dimensions, comparable to histopathological reconstructions.

Outcomes and Implications

The article suggests that this diagnostic method could be reliable in clinical practice even without human input. As early-stage melanoma is more curable than late-stage, this diagnostic method could increase the chances of early detection, which would increase chances of recovery. Additionally, as a noninvasive method of diagnosis, it would pose a decreased risk to the patient and reevaluation would be possible if necessary. The authors note the potential for this model to be applied in a clinical setting in the near future.