Multiple sclerosis and related disorders · 2026
Exploratory analysis of pupillography-based machine learning for assessing autonomic dysfunction in multiple sclerosis.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Parmak Yener Neslihan, Fırat Yelda, Seferoğlu Meral, Kargın Ahmet Metin, Kılıçaslan Yılmaz
Аннотация
BACKGROUND: The pupillary light reflex (PLR) is mediated by autonomic pathways, offering a potential noninvasive biomarker for Multiple sclerosis (MS). Pupillography enables objective quantification of PLR dynamics. However, its diagnostic utility in MS remains underexplored. This study aimed to investigate the feasibility of an ML-based analytical framework applied to pupillographic features rather than automated diagnosis.
METHODS: A total of 692 pupillographic images were obtained from 25 patients diagnosed with relapsing-remitting MS (RRMS) without optic neuritis (ON) and 38 age- and sex-matched healthy controls. For model development, data from 9 MS patients and 26 controls were used for training, and 16 MS patients and 12 controls for testing. Twenty-two quantitative features related to pupil size, shape, reflex dynamics, and symmetry were extracted. A random forest classifier was trained and evaluated using accuracy, sensitivity, specificity, and AUC-ROC. An independent validation set of 384 images from 18 MS patients and 14 controls was used to assess generalizability.
RESULTS: The model achieved 85.7% accuracy, 93.8% sensitivity, and an AUC-ROC of 0.945 on the test set, correctly identifying 15 of 16 MS cases. On the independent dataset, performance was lower (75.0% accuracy, 88.9% sensitivity, 57.1% specificity; AUC-ROC: 0.627). Features reflecting quadrant-based PLR variations contributed most to classification.
CONCLUSIONS: Pupillography-based ML models showed high sensitivity in distinguishing RRMS patients from healthy controls, may have potential as noninvasive adjunctive approaches. Although specificity decreased in the independent dataset, sensitivity remained relatively high. However, given the exploratory design and lack of disease controls, the findings should be interpreted with caution and require further validation.
METHODS: A total of 692 pupillographic images were obtained from 25 patients diagnosed with relapsing-remitting MS (RRMS) without optic neuritis (ON) and 38 age- and sex-matched healthy controls. For model development, data from 9 MS patients and 26 controls were used for training, and 16 MS patients and 12 controls for testing. Twenty-two quantitative features related to pupil size, shape, reflex dynamics, and symmetry were extracted. A random forest classifier was trained and evaluated using accuracy, sensitivity, specificity, and AUC-ROC. An independent validation set of 384 images from 18 MS patients and 14 controls was used to assess generalizability.
RESULTS: The model achieved 85.7% accuracy, 93.8% sensitivity, and an AUC-ROC of 0.945 on the test set, correctly identifying 15 of 16 MS cases. On the independent dataset, performance was lower (75.0% accuracy, 88.9% sensitivity, 57.1% specificity; AUC-ROC: 0.627). Features reflecting quadrant-based PLR variations contributed most to classification.
CONCLUSIONS: Pupillography-based ML models showed high sensitivity in distinguishing RRMS patients from healthy controls, may have potential as noninvasive adjunctive approaches. Although specificity decreased in the independent dataset, sensitivity remained relatively high. However, given the exploratory design and lack of disease controls, the findings should be interpreted with caution and require further validation.