ABSTRACT
ANTECEDENTES: La clasificación automática de imágenes es una rama prometedora del aprendizaje automático (de sus siglas en inglés Machine Learning [ML]), y es una herramienta útil en el diagnóstico de cáncer de piel. Sin embargo, poco se ha estudiado acerca de las limitaciones de su uso en la práctica clínica diaria. OBJETIVO: Determinar las limitaciones que existen en cuanto a la selección de imágenes usadas para el análisis por ML de las neoplasias cutáneas, en particular del melanoma. MÉTODOS: Se diseñó un estudio de cohorte retrospectivo, donde se incluyeron de forma consecutiva 2.849 imágenes dermatoscópicas de alta calidad de tumores cutáneos para su valoración por un sistema de ML, recogidas entre los años 2010 y 2014. Cada imagen dermatoscópica fue clasificada según las características de elegibilidad para el análisis por ML. RESULTADOS: De las 2.849 imágenes elegidas a partir de nuestra base de datos, 968 (34%) cumplieron los criterios de inclusión. De los 528 melanomas, 335 (63,4%) fueron excluidos. La ausencia de piel normal circundante (40,5% de todos los melanomas de nuestra base de datos) y la ausencia de pigmentación (14,2%) fueron las causas más frecuentes de exclusión para el análisis por ML. DISCUSIÓN: Solo el 36,6% de nuestros melanomas se consideraron aceptables para el análisis por sistemas de ML de última generación. Concluimos que los futuros sistemas de ML deberán ser entrenados a partir de bases de datos más grandes que incluyan imágenes representativas de la práctica clínica habitual. Afortunadamente, muchas de estas limitaciones están siendo superadas gracias a los avances realizados recientemente por la comunidad científica, como se ha demostrado en trabajos recientes
BACKGROUND: Automated image classification is a promising branch of machine learning (ML) useful for skin cancer diagnosis, but little has been determined about its limitations for general usability in current clinical practice. OBJECTIVE: To determine limitations in the selection of skin cancer images for ML analysis, particularly in melanoma. METHODS: Retrospective cohort study design, including 2,849 consecutive high-quality dermoscopy images of skin tumors from 2010 to 2014, for evaluation by a ML system. Each dermoscopy image was assorted according to its eligibility for ML analysis. RESULTS: Of the 2,849 images chosen from our database, 968 (34%) met the inclusion criteria for analysis by the ML system. Only 64.7% of nevi and 36.6% of melanoma met the inclusion criteria. Of the 528 melanomas, 335 (63.4%) were excluded. An absence of normal surrounding skin (40.5% of all melanomas from our database) and absence of pigmentation (14.2%) were the most common reasons for exclusion from ML analysis. DISCUSSION: Only 36.6% of our melanomas were admissible for analysis by state-of-the-art ML systems. We conclude that future ML systems should be trained on larger datasets which include relevant non-ideal images from lesions evaluated in real clinical practice. Fortunately, many of these limitations are being overcome by the scientific community as recent works show
Subject(s)
Humans , Melanoma/diagnosis , Skin Neoplasms/diagnosis , Diagnostic Imaging , Learning , Retrospective Studies , Cohort StudiesABSTRACT
BACKGROUND: Automated image classification is a promising branch of machine learning (ML) useful for skin cancer diagnosis, but little has been determined about its limitations for general usability in current clinical practice. OBJECTIVE: To determine limitations in the selection of skin cancer images for ML analysis, particularly in melanoma. METHODS: Retrospective cohort study design, including 2,849 consecutive high-quality dermoscopy images of skin tumors from 2010 to 2014, for evaluation by a ML system. Each dermoscopy image was assorted according to its eligibility for ML analysis. RESULTS: Of the 2,849 images chosen from our database, 968 (34%) met the inclusion criteria for analysis by the ML system. Only 64.7% of nevi and 36.6% of melanoma met the inclusion criteria. Of the 528 melanomas, 335 (63.4%) were excluded. An absence of normal surrounding skin (40.5% of all melanomas from our database) and absence of pigmentation (14.2%) were the most common reasons for exclusion from ML analysis. DISCUSSION: Only 36.6% of our melanomas were admissible for analysis by state-of-the-art ML systems. We conclude that future ML systems should be trained on larger datasets which include relevant non-ideal images from lesions evaluated in real clinical practice. Fortunately, many of these limitations are being overcome by the scientific community as recent works show.