||1R21CA240254-01A1 Interpret this number
||Sensory Cue Integration in Melanoma Screening
Imaging biomarkers are features in images that have biological implications. For example, in a picture of a person with
red hair, the red hair is a feature and the implication is that there is a mutation in the MC1R gene that provides instructions
for making a protein called the melanocortin 1 receptor. This feature, an imaging biomarker, can be used as a medical cue
to indicate increased risk for melanoma. When used in this context, this imaging biomarker becomes an imaging
biomarker cue (IBC), in the sense that it may cue the medical professional observer to alter treatment accordingly, such as
recommending sunscreen use. IBCs do not individually bear the full weight of medical decision-making and instead are
integrated. IBC analysis may be a process of sensory cue integration or may be a process of observation and integration
by technology such as a digital camera and computer. An advantage of the latter is that computational scalability enables
machine vision to compute vast permutations of IBCs that would be overwhelming to a human observer. Thus computers
can try many potential diagnostic methods rapidly before picking the best one to teach back to humans. The purpose of
this project is to develop a human/machine interface for bi-directional teaching so expert dermatologists can teach
computers what IBCs they use to achieve accurate diagnosis and computers can teach dermatologists the best way to use
current IBCs and suggest integration of new IBCs that machine learning guides them to. As an outcome, we will measure
the diagnostic performance of dermatologists who undergo IBC training in detecting melanoma. It is known that early
detection saves lives, but the potential of technology to improve early detection, a great need since 10,000 Americans still
die each year from melanoma, is unknown. This project will help answer that unknown and if we are successful in
translating IBCs with commuter vision and machine learning, more melanomas will be detected early and lives will be
saved. Our long-term goal is to reduce melanoma related deaths and unnecessary biopsies by helping clinicians increase
the predictive value of dermoscopy-based melanoma screening. We believe sensitivity and specificity of dermoscopy-
based melanoma screening for non-expert screeners can be improved by assistive technology, which is highly desirable
given the cost of false positives (patient stress and unnecessary biopsies) and the extremely high cost of false negatives
(delayed melanoma treatment).
The erythema Q-score, an imaging biomarker for redness in skin inflammation.
, Penzi L.
, Suarez-Farinas M.
, Garcet S.
, Brunner P.M.
, Czarnowicki T.
, Kim J.
, Bottomley C.
, Finney R.
, Cueto I.
, et al.
Experimental dermatology, 2021 Mar; 30(3), p. 377-383.
Deep learning-level melanoma detection by interpretable machine learning and imaging biomarker cues.
, Browning J.
, Correa Da Rosa J.
, Suarez-Farinas M.
, Lish S.
, Zong A.M.
, Firester B.
, Vrattos C.
, Renert-Yuval Y.
, Gamboa M.
, et al.
Journal of biomedical optics, 2020 Nov; 25(11), .