PREDICT: Precision Diagnostics for Early Melanoma Detection Using Spatial Biology and AI-Guided Image Analysis
Abstract
The rapid increase in suspected melanoma biopsies that require pathological tests is a significant burden to the doctors and the health care system. However, early diagnosis of melanoma is challenging, because malignant melanomas can mimic benign lesions that are overwhelmingly more common in the population. The ability to test suspected melanoma lesions in a fast, cost-effective, and accurate way is an essential, unmet need. Nevi (or moles) are commonly benign, but approximately 30% of all melanomas arise from a pre-existing nevus. If an incorrect benign diagnosis is made, no further curative surgery is performed for years, which may lead to a fatal metastasis. At the other end of the spectrum, an incorrect diagnosis of malignant may be provided to an otherwise banal nevus. Both scenarios have serious consequences. An incorrect melanoma diagnosis has long-lasting psychological effects, alters one’s health insurance risk, leads to increased healthcare costs for the patient and the health system, and incorrectly inflates the annual melanoma rates. Hematoxylin and Eosin (H&E) staining of tissue samples has been used by pathologists for more than a century. However, microscopic examination of H&E images is time-consuming for difficult cases, variable, and is not sensitive enough to define all cell types and areas of malignant potential with a high degree of accuracy. The inaccuracy and inconsistency are largely due to non-specific staining procedures combined with a pathologist’s best judgement of ever-changing morphological features. A nevus may be described as dysplastic by a pathologist, which means it does not visually appear normal or benign, but it also does not have the usual characteristics of a melanoma. There is very little agreeance among pathologists when the same tissue section is reviewed. There has also been a rapid rise of diagnosis of melanoma in situ (MIS) which is the non-invasive version. There are various reasons for this, including diagnostic uncertainty. Due to medico-legal reasons pathologists tend to err on the side of caution, rather the missing a potential malignant lesion. The proposed project will address this diagnostic uncertainty by using cutting-edge molecular analysis and imaging tools that will enable each cell-type present in the tissue to be identified and labelled with a unique gene expression signature. Our team also has developed software that enables the matching of these labelled cells with standard pathology stained images. With these tools we will be able to train our software to recognize areas of malignant potential within a standard scanned pathology slides. An example workflow to integrate our computer-assisted diagnostic tool into clinical practice: Pathology tissues are sectioned and stained using H&E (standard procedure), then scanned using current high-resolution slide scanners. The stained slide is diagnosed as per normal, with notes of observations taken by a Dictaphone. The H&E image is simultaneously analyzed by our computer-assisted tool to identify regions of malignancy or malignant potential, and a diagnostic report is generated. This rapid assessment is then cross validated with standard histopathological diagnosis. If discordance is observed, this will prompt subsequent re-review by additional pathologists to reach a consensus, before a final report is sent to the treating clinician for treatment/follow-up recommendations. Our computer-assisted tool will guide the need for further surgery of early lesions. In being able to re-classify an ambiguous lesion as either malignant or benign, our test will prompt the appropriate use of additional excisions, thereby halting further formation or spread of the early melanoma, while simultaneously preventing unnecessary additional surgery/treatment where lesions show no indication of malignant potential. If an additional wide local excision is required, our computer-assisted tool w
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 28, 2022
- Source ID
- W81XWH2210766
Entities
People
- Mitchell S. Stark
Organizations
- United States Army
- University of Queensland