lrsoenksen / SPL_UD_DL

A reported 96,480 people were diagnosed with melanoma in the United States in 2019, leading to 7230 reported deaths. Early-stage identification of suspicious pigmented lesions (SPLs) in primary care settings can lead to im- proved melanoma prognosis and a possible 20-fold reduction in treatment cost. Despite this clinical and economic value, efficient tools for SPL detection are mostly absent. To bridge this gap, we developed an SPL analysis system for wide-field images using deep convolutional neural networks (DCNNs) and applied it to a 38,283 dermatological dataset collected from 133 patients and publicly available images. These images were obtained from a variety of consumer-grade cameras (15,244 nondermoscopy) and classified by three board-certified dermatologists. Our system achieved more than 90.3% sensitivity (95% confidence interval, 90 to 90.6) and 89.9% specificity (89.6 to 90.2%) in distinguishing SPLs from nonsuspicious lesions, skin, and complex backgrounds, avoiding the need for cumbersome individual lesion imaging. We also present a new method to extract intrapatient lesion saliency (ugly duckling criteria) on the basis of DCNN features from detected lesions. This saliency ranking was validated against three board-certified dermatologists using a set of 135 individual wide-field images from 68 dermatolog- ical patients not included in the DCNN training set, exhibiting 82.96% (67.88 to 88.26%) agreement with at least one of the top three lesions in the dermatological consensus ranking. This method could allow for rapid and accurate assessments of pigmented lesion suspiciousness within a primary care visit and could enable improved patient triaging, utilization of resources, and earlier treatment of melanoma.

Deep Learning for Dermatologist-Level Detection of Ugly-Duckling (UD) and Suspicious Pigmented Skin Lesions (SPL) from Wide-Field Images

Code to reproduce Soenksen, LR. et al 2021, on Science Translational Medicine.

CODE STRUCTURE (NOTEBOOKS / INPUTS / OUTPUTS)

Samples of data Preparation, model training, testing and integrated analysis system according to the methods of Soenksen, LR. et al 2020, can be executed through the included Jupyter notebooks in the following order:

• 00_A_DL_Image_Patch_generation.ipynb
• 00_B_DL_Image_Database_CLAHE_PreProcessing.ipynb
• 00_C_DL_Image_Database_Randomization.ipynb
• 00_D_DL_Image_Augmentation_of_Randomized_CLAHE_Database.ipynb
• 01_DL_SPL_Detection_Basic_Model_Creator.ipynb
• 02_DL_SPL_Detection_Augmented_Model_Creator.ipynb
• 03_DL_SPL_Detection_Augmented_TL_VGG16_Bottleneck_Model_Creator.ipynb
• 03_DL_SPL_Detection_Augmented_TL_VGG16_Fine_Tuning_Model_Creator.ipynb
• 04_DL_SPL_Detection_Augmented_TL_XCEPTION_Bottleneck_Model_Creator.ipynb
• 04_DL_SPL_Detection_Augmented_TL_XCEPTION_Fine_Tuning_Model_Creator.ipynb
• 05_DL_SPL_A_Wide_Field_Feature_Extractor_UglyDucking_Ranking_and_T-SNE.ipynb

PROBLEM/SOLUTION DEFINITION

Wide-field imaging and deep neural networks are used to facilitate the accurate detection of suspicious and salient pigmented lesions to allow for convenient skin screenings at the primary care level.

DATASET (Direct Download)

All data has been de-identified and randomized to comply with MIT data sharing policies. Due to egress limits on GIT, this repo requires that you download the "Wide-field" and "close-up" Dataset directly. Please send a request to [e-mail] with your Name, Organization, Position, and Anticipated project aims. We will try to respond as quickly as we can to provide you with a secure link to access it. Data and code are provided ONLY for non-commertial purposes. After download place in the main project folder to use with the provided code.

MODELS (Direct Download)

Due to egress limits on GIT, this repo requires that you download the following "Outputs" folder, which includes the trained Deep Convolutional Neural Network (DCNN) model weight files directly from this link: https://www.dropbox.com/s/bfjqv5yfynxr6sd/Models.zip?dl=0. After download place in the main project folder and unzip.