A method for enhancing the performance of computer vision programs based on artificial intelligence models using a correction module for photo image parameters

Background. A previous analysis of the causes of misclassification of melanocytic and non-melanocytic skin lesions by computer vision programs (CVPs) based on the artificial intelligence (AI) models Derma Onko Check and Melanoma Check revealed systematic deviations in image quality metrics within groups of false-positive and false-negative results. Elimination of these deviations through targeted correction of photo image parameters is a logical step toward improving the diagnostic accuracy of AI-based systems.
Objective: To develop a module for correcting photo image parameters, which is capable of normalizing quality metrics to the reference ranges of correctly classified cases, as well as to conduct a quantitative assessment of its impact on the performance of CVPs.
Material and methods. Photographic images from an anonymized skin lesion database were subjected to parameter correction. A set of 13 photometric and texture metrics was calculated for each image; the characteristics of correctly classified photo images (true positive and true negative cases) were used as normal ranges. A Python module was developed that implements sequential, independent correction of the following deviating parameters: white balance, gamma correction, adaptive contrast processing, and an unsharp mask. Re-inference of the processed photo images was performed using Tensor Flow Lite (TFLite) models with routing to the corresponding programs Derma Onko Check and Melanoma Check. Performance was assessed by accuracy, sensitivity, and specificity.
Results. The developed correction module improved the diagnostic accuracy of AI-based CVPs. During validation of the AI program Derma Onko Check on the dataset of melanocytic neoplasms (n=230), the accuracy increased from 0.909 to 0.926 (+0.017), sensitivity – from 0.949 to 0.961 (+0.012), specificity – from 0.901 to 0.919 (+0.019). During validation of the AI program Melanoma Check on the dataset of melanocytic neoplasms, the accuracy increased from 0.844 to 0.857 (+0.014). During validation of the AI program Derma Onko Check on the dataset of non-melanocytic neoplasms (n=230), the accuracy increased from 0.868 to 0.882 (+0.015). Images with critical defects (overexposed pixels exceeding 55%) were excluded from the inference: seven images for the Derma Onko Check AI program and five images for the Melanoma Check AI program in the melanocytic dataset. No critically defective images were identified in the non-melanocytic tumor dataset.
Conclusion. The developed module for correcting image parameters ensures a stable and reproducible improvement in the diagnostic accuracy (sensitivity and specificity) of CVPs. The obtained results confirm the feasibility of integrating such a module into CVPs.

1. Zbrzezny A.M., Krzywicki T. Artificial intelligence in dermatology: a review of methods, clinical applications, and perspectives. Appl Sci. 2025; 15 (14): 7856. https://doi.org/10.3390/app15147856.

2. Olveres J., González G., Torres F., et al. What is new in computer vision and artificial intelligence in medical image analysis applications. Quant Imaging Med Surg. 2021; 11 (8): 3830–53. https://doi.org/10.21037/qims-20-1151.

3. Lamotkin A.I., Korabelnikov D.I., Lamotkin I.A., et al. Artificial intelligence in healthcare and medicine: the history of key events, its significance for doctors, the level of development in different countries. FARMAKOEKONOMIKA. Sovremennaya farmakoekonomika i farmakoepidemiologiya / FARMAKOEKONOMIKA. Modern Pharmacoeconomics and Pharmacoepidemiology. 2024; 17 (2): 243–50 (in Russ.). https://doi.org/10.17749/2070-4909/farmakoekonomika.2024.254.

4. Korabelnikov D.I., Lamotkin A.I. The effectiveness of using artificial intelligence in clinical medicine. FARMAKOEKONOMIKA. Sovremennaya farmakoekonomika i farmakoepidemiologiya / FARMAKOEKONOMIKA. Modern Pharmacoeconomics and Pharmacoepidemiology. 2025; 18 (1): 114–24 (in Russ.). https://doi.org/10.17749/2070-4909/farmakoekonomika.2025.287.

5. Goyal M., Reeves N.D., Davison A.K., et al. Artificial intelligencebased image classification methods for diagnosis of skin cancer: challenges and opportunities. Comput Biol Med. 2020; 127: 104065. https://doi.org/10.1016/j.compbiomed.2020.104065.

6. Han S.S., Moon I.J., Kim S.H., et al. The degradation of performance of a state-of-the-art skin image classifier when applied to patient-driven internet search. Sci Rep. 2022; 12 (1): 16270. https://doi.org/10.1038/s41598-022-20632-7.

7. Korabelnikov D.I., Lamotkin A.I. Analysis of the reasons of misclassification of non-melanocytic skin tumors by artificial intelligence based programs. Siberian Journal of Clinical and Experimental Medicine. 2026; 41 (1): 221–31 (in Russ.). https://doi.org/10.29001/2073-8552-2026-2881.

8. Han S.S., Moon I.J., Lim W., et al. Keratinocytic skin cancer detection on the face using region-based convolutional neural network. JAMA Dermatol. 2020; 156 (1): 29–37. https://doi.org/10.1001/jamadermatol.2019.3807.

9. Vodrahalli K., Daneshjou R., Novoa R.A., et al. TrueImage: a machine learning algorithm to improve the quality of telehealth photos. Biocomputing 2021. November 2020: 220–31. https://doi.org/10.1142/9789811232701_0021.

10. Li Z., Koban K.C., Schenck T.L., et al. Artificial intelligence in dermatology image analysis: current developments and future trends. J Clin Med. 2022; 11 (22): 6826. https://doi.org/10.3390/jcm11226826.