Uso del aprendizaje automático en el diagnóstico del melanoma. Limitaciones por superar / Machine Learning in Melanoma Diagnosis. Limitations About to be Overcome

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

Machine Learning in Melanoma Diagnosis. Limitations About to be Overcome. / Uso del aprendizaje automático en el diagnóstico del melanoma. Limitaciones por superar.

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.

In vivo reflectance confocal microscopy of shave biopsy wounds: feasibility of intraoperative mapping of cancer margins.

BACKGROUND: Reflectance confocal microscopy (RCM) images skin at cellular resolution and has shown utility for the diagnosis of nonmelanoma skin cancer in vivo. Topical application of aluminium chloride (AlCl(3)) enhances contrast in RCM images by brightening nuclei. OBJECTIVES: To investigate feasibility of RCM imaging of shave biopsy wounds using AlCl(3) as a contrast agent. METHODS: AlCl(3) staining was optimized, in terms of concentration vs. immersion time, on excised tissue ex vivo. RCM imaging protocol was tested in patients undergoing shave biopsies. The RCM images were retrospectively analysed and compared with the corresponding histopathology. RESULTS: For 35% AlCl(3) , routinely used for haemostasis in clinic, minimum immersion time was determined to be 1 min. We identified three consistent patterns of margins on RCM mosaic images by varying depth: epidermal margins, peripheral dermal margins, and deep dermal margins. Tumour islands of basal cell carcinoma were identified at peripheral or deep dermal margins, correlating on histopathology with aggregates of neoplastic basaloid cells. Atypical cobblestone or honeycomb patterns were identified at the epidermal margins in squamous cell carcinomas, correlating with a proliferation of atypical keratinocytes extending to biopsy margins. CONCLUSIONS: RCM imaging of shave biopsy wounds is feasible and demonstrates the future possibility of intraoperative mapping in surgical wounds.

Detection of basal cell carcinomas in Mohs excisions with fluorescence confocal mosaicing microscopy.

BACKGROUND: High-resolution real-time imaging of human skin is possible with a confocal microscope either in vivo or in freshly excised tissue ex vivo. Nuclear and cellular morphology is observed in thin optical sections, similar to that in conventional histology. Contrast agents such as acridine orange in fluorescence and acetic acid in reflectance have been used in ex vivo imaging to enhance nuclear contrast. OBJECTIVES: To evaluate the sensitivity and specificity of ex vivo real-time imaging with fluorescence confocal mosaicing microscopy, using acridine orange, for the detection of residual basal cell carcinoma (BCC) in Mohs fresh tissue excisions. METHODS: Forty-eight discarded skin excisions were collected following completion of Mohs surgery, consisting of excisions with and without residual BCC of all major subtypes. The tissue was stained with acridine orange and imaged with a fluorescent confocal mosaicing microscope. Confocal mosaics were matched to the corresponding haematoxylin and eosin-stained Mohs frozen sections. Each mosaic was divided into subsections, resulting in 149 submosaics for study. Two Mohs surgeons, who were blinded to the cases, independently assessed confocal submosaics and recorded the presence or absence of BCC, location, and histological subtype(s). Assessment of confocal mosaics was by comparison with corresponding Mohs surgery maps. RESULTS: The overall sensitivity and specificity of detecting residual BCC was 96.6% and 89.2%, respectively. The positive predictive value was 92.3% and the negative predictive value 94.7%. Very good correlation was observed between confocal mosaics and matched Mohs frozen sections for benign and malignant skin structures, overall tumour burden and location, and identification of all major histological subtypes of BCC. CONCLUSIONS: Fluorescent confocal mosaicing microscopy using acridine orange enables detection of residual BCC of all subtypes in Mohs fresh tissue excisions with high accuracy. This observation is an important step towards the long-term clinical goal of using a noninvasive imaging modality for potential real-time surgical pathology-at-the-bedside for skin and other tissues.

Confocal mosaicing microscopy in skin excisions: a demonstration of rapid surgical pathology.

Precise micro-surgical removal of tumour with minimal damage to the surrounding normal tissue requires a series of excisions, each guided by an examination of frozen histology of the previous. An example is Mohs surgery for the removal of basal cell carcinomas (BCCs) in skin. The preparation of frozen histology is labour-intensive and slow. Confocal microscopy may enable rapid detection of tumours directly in surgical excisions with minimal need for frozen histology. Mosaicing of images enables observation of nuclear and cellular morphology in large areas of surgically excised tissue. In skin, the use of 10-1% acetic acid as a reflectance contrast agent brightens nuclei in 0.5-5 min and enhances nuclear-to-dermis contrast and detectability of BCCs. A tissue fixture was engineered for precisely mounting surgical excisions to enable mosaicing of 36 x 36 images to create a field of view of 12 x 12 mm. This large field of view displays the excision at 2x magnification, similar to that routinely used by Mohs surgeons when examining frozen histology. Comparison of mosaics to histology demonstrates detectability of BCCs. Confocal mosaicing presently requires 9 min, instead of 20-45 min per excision for preparing frozen histology, and thus may provide a means for rapid pathology-at-the-bedside to expedite and guide surgery.

Muscle control in chronic tic disorders.

EMG was recorded in nine subjects suffering from chronic tic disorder. Six subjects suffered asymmetrical tics and three had symmetrical tics. EMG in tic-affected and contralateral nonaffected sites was recorded at rest, during a baseline period, and at postbiofeedback training. All subjects received 2-4 biofeedback training sessions aimed at enhancing their ability to control levels of muscle contraction in both affected and nonaffected sites. All nine subjects met the criterion of discriminating unaided between levels of 0, 25%, 50%, and 75% of their fullest contraction. Five of the six people with asymmetrical tics showed lower resting EMG on the affected side at baseline, but EMG significantly increased in tic-affected but not nonaffected muscles after exercises aimed at enhancing muscle control. Six subjects reported a clinically significant > or = 40% decrease in tic frequency. The reflexlike quality of tic muscles can modified by biofeedback training and this constitutes a useful and relatively quickly acquired aid to tic management.

Personal constructs associated with tics.

Thirteen subjects experiencing chronic tics kept diaries noting frequency, intensity and degree of control of their tic during their daily routines for one week. Three examples of high-, medium- and low-risk situations were chosen as elements and, using a modified form of Kelly's repertory grid, bipolar constructs were elicited by comparing the three types of situations. Principal dimensions extracted from INGRID analyses differed across individuals but related to three principal areas; self-image, degree of task involvement and type of task demand. Feelings of impatience and frustration frequently accompanied tic onset. Subjects' cognitive evaluations of their tic situations may be important in tic management.

Changes in construals of tic-producing situations following cognitive and behavioral therapy.

12 clients suffering from chronic tics participated in one of two treatment programs, either a behavioral group using competing response therapy or a group using Beck-style cognitive restructuring. A repertory grid based upon the personal construct psychology of George Kelly was administered to all clients before and after treatment. The grid comprised a set of elements made up of situations with high, medium, and low risk of eliciting tics, and constructs were derived from comparisons between them. Clients' ratings of the elements on the constructs were subjected to a principal components analysis using an INGRID program. Following treatment the total variation around construct means decreased in both groups but significantly more in the cognitive group, indicating a narrowing of the difference in their perceptions of situations which formerly indicated high and low risk of inducing tics.