Dataset Information

A deep-learning toolkit for visualization and interpretation of segmented medical images.

ABSTRACT: Generalizability of deep-learning (DL) model performance is not well understood and uses anecdotal assumptions for increasing training data to improve segmentation of medical images. We report statistical methods for visual interpretation of DL models trained using ImageNet initialization with natural-world (T _II) and supervised learning with medical images (L _MI) for binary segmentation of skin cancer, prostate tumors, and kidneys. An algorithm for computation of Dice scores from union and intersections of individual output masks was developed for synergistic segmentation by T _II and L _MI models. Stress testing with non-Gaussian distributions of infrequent clinical labels and images showed that sparsity of natural-world and domain medical images can counterintuitively reduce type I and type II errors of DL models. A toolkit of 30 T _II and L _MI models, code, and visual outputs of 59,967 images is shared to identify the target and non-target medical image pixels and clinical labels to explain the performance of DL models.

SUBMITTER: Ghosal S

PROVIDER: S-EPMC9017181 | biostudies-literature | 2021 Nov

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

A deep-learning toolkit for visualization and interpretation of segmented medical images.

Ghosal Sambuddha S Shah Pratik P

Cell reports methods 20211108 7

Generalizability of deep-learning (DL) model performance is not well understood and uses anecdotal assumptions for increasing training data to improve segmentation of medical images. We report statistical methods for visual interpretation of DL models trained using ImageNet initialization with natural-world (T II) and supervised learning with medical images (L MI) for binary segmentation of skin cancer, prostate tumors, and kidneys. An algorithm for computatio ...[more]

PMID: 35474999

Similar Datasets

Project description:BackgroundA change in the output of deep neural networks (DNNs) via the perturbation of a few pixels of an image is referred to as an adversarial attack, and these perturbed images are known as adversarial samples. This study examined strategies for compromising the integrity of DNNs under stringent conditions, specifically by inducing the misclassification of medical images of disease with minimal pixel modifications.MethodsThis study used the following three publicly available datasets: the chest radiograph of emphysema (cxr) dataset, the melanocytic lesion (derm) dataset, and the Kaggle diabetic retinopathy (dr) dataset. To attack the medical images, we proposed a method termed decrease group differential evolution (DGDE) for generating adversarial images. Under this method, a noise matrix of the same size as the input image is first used to attack the image sample several times until the initial adversarial perturbation s0 is obtained. Next, a subset s1 is randomly picked from the initial adversarial perturbation s0 that is still able to cause the samples to be misclassified by the classifier. A new subset s2 is subsequently randomly selected from the adversarial perturbation subset s1, which can still cause the adversarial samples to be misclassified by the classifier. Finally, the adversarial perturbation subset sn with the minimum number of elements is obtained by continuous reduction of the number of perturbed pixels.ResultsIn this study, the DGDE method was used to attack the images of the cxr dataset, the derm dataset, and the dr dataset; and the minimum number of pixels required to be considered a successful attack was 11, 7, and 7, respectively, while the maximum number of pixel changes was 55, 35, and 21, respectively. The average number of pixel changes was 30, 18, and 11, respectively, in the cxr dataset, the derm dataset, and the dr dataset, respectively, while the percentages of the average number of pixel changes among the total number of pixels of the image were 0.0598%, 0.0359%, and 0.0219, respectively.ConclusionsUnlike the traditional differential evolution (DE) method, the proposed DGDE method modifies fewer pixels to generate adversarial samples by introducing a variable population number and a novel crossover and selection strategy. However, the success rate of the initial attack on different image datasets varied greatly. In future studies, we intend to identify the reasons for this phenomenon and improve the success rate of the initial attack.

Project description:PurposeThis study aimed to develop a fully automated deep learning ciliary body segmentation and assessment approach in three-dimensional ultrasound biomicroscopy (3D-UBM) images.MethodsEach 3D-UBM eye volume was aligned to the optic axis via multiplanar reformatting. Ciliary muscle and processes were manually annotated, and Deeplab-v3+ models with different loss functions were trained to segment the ciliary body (ciliary muscle and processes) in both en face and radial images.ResultsWe trained and tested the models on 4320 radial and 3864 en face images from 12 cadaver eye volumes. Deep learning models trained on radial images with Dice loss achieved the highest mean F1-score (0.89) for ciliary body segmentation. For three-class segmentation (ciliary muscle, processes, and background), radial images with Dice loss achieved the highest mean F1-score (0.75 for the ciliary process and 0.82 for the ciliary muscle). Part of the ciliary muscle (10.9%) was misclassified as the ciliary process and vice versa, which occurred owing to the difficulty in differentiating the ciliary muscle-processes border, even by experts. Deep learning segmentation made further editing by experts at least seven times faster than a fully manual approach. In eight cadaver eyes, the average ciliary muscle, process, and body volumes were 56 ± 9, 43 ± 13, and 99 ± 18 mm3, respectively. The average surface area of the ciliary muscle, process, and body were 346 ± 45, 363 ± 83, and 709 ± 80 mm2, respectively. We performed transscleral cyclophotocoagulation in cadaver eyes to shrink the ciliary processes. Both manual and automated measurements from deep learning segmentation show a decrease in volume, surface area, and 360° cross-sectional area measurements.ConclusionsThe proposed deep learning segmentation of the ciliary body and 3D measurements showed transscleral cyclophotocoagulation-related changes in the ciliary body.Translational relevanceAutomated ciliary body assessment using 3D-UBM has the translational potential for ophthalmic treatment planning and monitoring.

Dataset Information

A deep-learning toolkit for visualization and interpretation of segmented medical images.

Publications

A deep-learning toolkit for visualization and interpretation of segmented medical images.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets