Deep learning approach to skin layers segmentation in inflammatory dermatoses

Czajkowska, Joanna; Badura, Paweł; Korzekwa, Szymon; Płatkowska-Szczerek, Anna

doi:10.1016/j.ultras.2021.106412

Cited by 17 publications

(40 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The first one, further called benchmark, consists of 380 HFUS images of 380 patients with atopic dermatitis (303 images) or psoriasis (77). The dataset is publicly available [ 37 ] along with the pre-trained SegUnet model for skin layer segmentation [ 7 ]. The second dataset consists of 200 images of 32 patients with different non-melanocytic skin tumors: BCC (143 images), fibroma (32), skin metastasis of breast cancer (10), keratofibroma (9), superficial BCC (3), and squamous cell carcinoma (3), whereas the third data set includes 51 images of 51 patients with healthy skin.…”

Section: Materials and Methodsmentioning

confidence: 99%

“…US at frequencies below 100 MHz is now commonly used in medical practice [ 3 ], and the >100 MHz probes are constantly being designed [ 4 ]. Improving spatial resolution of acquired images, the higher frequency of the US probe enables clear visualization of superficial structures such as the fat layer, the muscle layer, blood vessels, hair follicles, and skin appendages [ 2 , 5 , 6 , 7 ]. HFUS is used for healthy skin analysis, where skin thickness is inversely proportional to age, and due to increased collagen production connected with aging, echogenicity tends to increase too [ 2 ].…”

Section: Introductionmentioning

confidence: 99%

“…The presence of a subepidermal low echogenic band (SLEB) in visually healthy skin enables skin lesion differentiation. Its thickness correlates with the histological degree of epidermal hyperkeratosis and intensity of inflammatory infiltrates [ 2 , 7 , 10 , 11 ]. Although different studies [ 2 , 11 ] examine the validity, repeatability, and reliability of skin measurements, which correlate with histological analysis, none of them analyze the HFUS capability in skin lesion differentiation.…”

Section: Introductionmentioning

confidence: 99%

“…Despite the lack of algorithms for HFUS image classification, different methods have appeared for automated and semi-automated HFUS (or UHFUS) image analysis. Pereyra et al [ 12 ] first described the segmentation of skin lesions in a 25 MHz US using a generalized Rayleigh mixture model, then References [ 7 , 13 , 14 , 15 , 16 ] presented automatic skin layer segmentation algorithms. They employed classic segmentation approaches such as level sets and active contour models [ 14 , 15 , 16 ], or the most recent deep learning-based solutions [ 7 ].…”

Section: Introductionmentioning

confidence: 99%

“…Pereyra et al [ 12 ] first described the segmentation of skin lesions in a 25 MHz US using a generalized Rayleigh mixture model, then References [ 7 , 13 , 14 , 15 , 16 ] presented automatic skin layer segmentation algorithms. They employed classic segmentation approaches such as level sets and active contour models [ 14 , 15 , 16 ], or the most recent deep learning-based solutions [ 7 ]. The latter is now widely used in image segmentation, mainly through convolutional neural networks (CNNs), consisting of different layers responsible for dedicated data processing [ 17 ].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Deep Learning-Based High-Frequency Ultrasound Skin Image Classification with Multicriteria Model Evaluation

Czajkowska

Badura

Korzekwa

et al. 2021

Sensors

Self Cite

View full text Add to dashboard Cite

This study presents the first application of convolutional neural networks to high-frequency ultrasound skin image classification. This type of imaging opens up new opportunities in dermatology, showing inflammatory diseases such as atopic dermatitis, psoriasis, or skin lesions. We collected a database of 631 images with healthy skin and different skin pathologies to train and assess all stages of the methodology. The proposed framework starts with the segmentation of the epidermal layer using a DeepLab v3+ model with a pre-trained Xception backbone. We employ transfer learning to train the segmentation model for two purposes: to extract the region of interest for classification and to prepare the skin layer map for classification confidence estimation. For classification, we train five models in different input data modes and data augmentation setups. We also introduce a classification confidence level to evaluate the deep model’s reliability. The measure combines our skin layer map with the heatmap produced by the Grad-CAM technique designed to indicate image regions used by the deep model to make a classification decision. Moreover, we propose a multicriteria model evaluation measure to select the optimal model in terms of classification accuracy, confidence, and test dataset size. The experiments described in the paper show that the DenseNet-201 model fed with the extracted region of interest produces the most reliable and accurate results.

show abstract

Section: Materials and Methodsmentioning

confidence: 99%