A Calibrated and Conformal Deep Learning Framework for Trustworthy Antinuclear Antibody Pattern Recognition with Selective Referral to Experts

Authors

• Hussein Ali Hussein Al Naffakh, Faculty of Computer Science and Information Technology, Universiti Tun Hussein Onn Malaysia (UTHM), Batu Pahat, MalaysiaFollow
• Ahmed Dheyaa Radhi, College of Pharmacy, University of Al-Ameed, Karbala PO Box 198, Iraq
• Raghdah Maytham Hameed, Department of Medical Laboratory Techniques, University of Alkafeel, Najaf, Iraq
• Muntaha Abdullah Reishaan, Department of Anesthesia, College of Health and Medical Technology, University of Alkafeel, Najaf, Iraq
• Fouad A. Majeed, Department of Physics, College of Education for Pure Sciences, University of Babylon, Babylon, Iraq
• Rozaida Ghazali, Faculty of Computer Science and Information Technology, Universiti Tun Hussein Onn Malaysia (UTHM), Batu Pahat, Malaysia

Abstract

Reading antinuclear antibody patterns on human epithelial cells by indirect immunofluorescence is the reference screen for systemic autoimmune rheumatic diseases, but it is slow, subjective, and variable between observers. Deep learning reaches high accuracy on this task, yet most systems return a single prediction without stating how reliable it is, which is unsafe in a diagnostic workflow. This paper presents an intelligent decision support framework built around a single calibrated uncertainty signal. That signal is the control variable for four reliability modules: confidence calibration, conformal prediction, error detection, and selective referral. A feature space out of distribution detector serves as a safeguard. A convolutional neural network was trained on a public collection of more than 63,000 cell images. It reached an accuracy of 94.8% and detected its own errors with an area under the curve of 0.947. Referring only the least confident 20% of cases raised the accuracy on the retained cases to 99.5%. These results reflect a cell level partition. Applied without retraining to an independent collection acquired at another laboratory, the same network reached 54.3%, while selective referral and error detection continued to work. The internal figure is therefore an upper bound on deployable performance, and the external figure a lower bound. Clinically, the system clears the confident majority of cells automatically and routes the difficult minority to an expert, which reduces reading workload and inter observer variability while preserving a human safety net. The framework transforms an accurate classifier into a trustworthy decision support system

Recommended Citation

Al Naffakh, Hussein Ali Hussein; Radhi, Ahmed Dheyaa; Hameed, Raghdah Maytham; Reishaan, Muntaha Abdullah; Majeed, Fouad A.; and Ghazali, Rozaida (2026) "A Calibrated and Conformal Deep Learning Framework for Trustworthy Antinuclear Antibody Pattern Recognition with Selective Referral to Experts," Karbala International Journal of Modern Science: Vol. 12 : Iss. 3 , Article 16.
Available at: https://doi.org/10.33640/2405-609X.3481

Creative Commons License

This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.