Deep Learning-Based Psoriasis Severity Staging

This paper has presented a structured survey of twenty deep learning studies on psoriasis image analysis published between 2017 and 2023. The reviewed works collectively demonstrate that transfer learning-based convolutional architectures can achieve clinically meaningful accuracy on severity staging and related tasks, with compound-scaled EfficientNet variants consistently outperforming earlier VGG and MobileNet designs under controlled comparisons. Four persistent structural gaps were identified: the underrepresentation of three-class severity staging relative to binary detection tasks; insufficient attention to the moderate-severity boundary where classification errors concentrate; limited standardised multi-institutional datasets; and the absence of universally adopted evaluation benchmarks enabling reliable cross-study comparison. These gaps collectively motivate a dedicated severity staging pipeline with psoriasis-specific preprocessing, lesion-aware augmentation targeting boundary-case coverage, and rigorous per-class benchmarking. Future extensions include integration of upstream lesion segmentation for spatial attention, mobile deployment, and longitudinal severity tracking within electronic health record systems.

[1]   A. Maulana et al., "Deep learning-based psoriasis severity classification using ResNet50, VGG19, MobileNetV3, and EfficientNetB0," J. Comput. Sci. Inf. Technol., vol. 10, no. 2, pp. 45-60, 2023. [2]  Y. Xing et al., "Deep learning framework for automated PASI score prediction using clinical trial images," IEEE J. Biomed. Health Inform., vol. 26, no. 4, pp. 1523-1532, Apr. 2022. [3]   S. F. Aijaz et al., "Deep learning for psoriasis subtype classification using neural network and LSTM architectures," Comput. Med. Imaging Graph., vol. 98, p. 102066, Jun. 2022. [4]  A. Goswami and N. Sharma, "YOLO-powered deep learning for automated psoriasis classification," in Proc. IEEE BIBM, Las Vegas, NV, USA, 2022, pp. 1432-1438. [5] C. Bolia and S. Joshi, "Hybrid EfficientNet-LSTM for psoriasis dermoscopic image classification," Expert Syst. Appl., vol. 215, p. 119409, 2023. [6]   G .-S. Lin et al., "Transfer learning for psoriasis image segmentation and analysis," Comput. Biol. Med., vol. 148, p. 105922, Sep. 2022. [7]  V. K. Semwal et al., "PsLSNet: Automated psoriasis lesion segmentation using modified U-Net," Biomed. Signal Process. Control, vol. 78, p. 103979, Aug. 2022. [8]   V. N. Shrivastava et al., "Machine learning-based psoriasis classification using image processing," J. Dermatol. Sci., vol. 89, pp. 112-120, 2021. [9]   J. Jung and C. Wu, "Psoriasis identification using k-means clustering and morphological processing," Comput. Biol. Med., vol. 132, p. 104297, 2021. [10]L. Amruthalingam et al., "Deep learning for quantification of pustular psoriasis lesions," J. Invest. Dermatol., vol. 141, no. 7,pp. 1812-1820, Jul. 2021.