Page 123 - AIH-1-4

Artificial Intelligence in Health Complex early diagnosis of MS through machine learning

Table 3. Top five feature interactions for tree‑based models 3.3.3. Interaction between gender and
Periventricular_MRI
Model Feature 1 Feature 2 Interaction
value Gender also has a significant interaction with
CatBoost Oligoclonal_Bands Periventricular_MRI 0.097347 Periventricular_MRI, as depicted in Figure 9. All
Gender Periventricular_MRI 0.078551 three models with the highest prediction performance
Oligoclonal_Bands Varicella 0.070430 (CatBoost, XGBoost, and LGBM) effectively capture this
Symptom_Sensory Symptom_Vision 0.067328 interaction, as indicated by the distinct clusters of colors
Periventricular_MRI VEP 0.057325 in the subplots. For females (Gender = 0), the interaction
XGBoost Oligoclonal_Bands Periventricular_MRI 0.131877 values clustered around zero, suggesting that MRI findings
Gender Periventricular_MRI 0.099022 have a more balanced impact. In contrast, males (Gender
= 1) showed more distinct interaction values, especially
Symptom_Motor ULSSEP 0.077204 when periventricular lesions were present, indicating that
Age Cortical_MRI 0.069749 the presence of these lesions in males has a greater impact
Mono_Symptomatic Symptom_Sensory 0.067236 on the prediction. This pattern suggests a potential gender-
LGBM Symptom_Motor ULSSEP 0.106938 related difference in how periventricular lesions influence
Oligoclonal_Bands Periventricular_MRI 0.063874 model predictions, which could have implications
Mono_Symptomatic Symptom_Sensory 0.062791 for understanding gender-specific progression or
Mono_Symptomatic Symptom_Vision 0.062664 manifestation of neurological conditions associated with
Breastfeeding Mono_Symptomatic 0.058049 these features.
RF Oligoclonal_Bands Periventricular_MRI 0.009170
Gender Periventricular_MRI 0.007391 3.3.4. Interaction between Symptom_Motor and
Infratentorial_MRI Oligoclonal_Bands 0.006605 ULSSEP
Infratentorial_MRI Periventricular_MRI 0.004550 Another significant pair of features is Symptom_Motor
Oligoclonal_Bands Varicella 0.004510 and ULSSEP, visualized in Figure 10. ULSSEP stands
Abbreviations: CatBoost: Categorical boosting; LGBM: Light gradient for upper limb somatosensory evoked potentials. It is a
boosting machine; RF: Random forest; XGBoost: Extreme gradient diagnostic test that measures the electrical activity in the
boosting. brain in response to stimulation of the sensory nerves in

Figure 8. Interactions of Oligoclonal_Bands and Periventricular_MRI on four machine learning models
Abbreviations: CatBoost: Categorical boosting; LGBM: Light gradient boosting machine; RF: Random forest; XGBoost: Extreme gradient boosting.

Volume 1 Issue 4 (2024) 117 doi: 10.36922/aih.4255