Can’t see the forest for the trees

Random forests are currently one of the most popular algorithms for supervised machine learning tasks. By taking into account for many trees instead of a single one the resulting forest model is no longer easy to understand and also often denoted as a black box. The paper is dedicated to the interpretability of random forest models using tree-based explanations. Two different concepts, namely most representative trees and surrogate trees are analyzed regarding both their ability to explain the model and to be understandable by humans. For this purpose explanation trees are further extended to groves, i.e. small forests of few trees. The results of an application to three real world data sets underline the inherent trade of between both requirements. Using groves allows to control for the complexity of an explanation while simultaneously analyzing their explanatory power.

The neurologic examination found a rapid erratic eye jerk corresponding to an opsoclonus, a gait ataxia, and a Romberg sign. No myoclonus was observed. Brain MRI was unremarkable, with no abnormal contrast enhancement. Two consecutive lumbar punctures found a slight pleocytosis with 15 and 55 leukocytes/mm 3 , respectively (mixed mono and polynuclear cells), with an elevated CSF protein level (1.6g/L), and negative bacteriologic direct and PCR assessments. Oral corticotherapy (1 mg/kg) was initiated regarding the assumption of carcinomatous meningitis or paraneoplastic encephalitis. The symptoms worsened within the following 2 weeks with inability to walk alone. Subsequent lumbar puncture found 149 leukocytes/mm 3 , with a large predominance of mononuclear cells (90%). Paraneoplastic antibodies were present neither in the serum nor in the CSF. Immunoglobulin M and immunoglobulin G antibodies against Borrelia burgdorferi were found in the blood with an ELISA (68 AU/mL [positive if > 14 AU/mL] and 119 AU/mL [positive if > 21 AU/mL], respectively) and then confirmed by a Western Blot analysis. A high rate of specific IgG antibodies was found in the CSF, 84.5 AU/mL (positive if > 5.4 AU/mL), reflecting an intrathecal synthesis with a positive index of specific IgG antibodies (Reiber formula). A diagnosis of neuroborreliosis was assumed, corticosteroids were stopped, and a 3-week treatment of ceftriaxone (2 g/d) was administrated. The patient showed an improvement after 1 week and a complete resolution of the symptoms after 4 weeks. Of note, no tick bite was reported, but the patient used to have hikes in forests in a French endemic area of Lyme disease.

Discussion
Opsoclonus is part of the opsoclonus-myoclonus ataxia syndrome (OMAS) which is, most of the time, a paraneoplastic syndrome associated with various tumors. Other causes include numerous infectious agents such as human immunodeficiency viruses, human herpesvirus 6, cytomegalovirus, Epstein-Barr virus, coxsackie virus B, mycoplasma, West Nile virus, and rubella. 1 Neuroborreliosis is a rare cause of opsoclonus with only 4 adults and 3 children cases reported in the literature. [2][3][4][5][6] Most patients displayed incomplete presentations, such as in our case who only presented with opsoclonus and gait ataxia, making the clinical diagnosis difficult, all the more so because the patient did not report erythema migrans.
Our case was also challenging because breast cancer is a frequent cause of paraneoplastic syndrome including OMAS. 1 This observation illustrates that neuroborreliosis, as well as others infections, should be considered as a differential diagnosis in OMAS even in apparently obvious situations, given the consequences of therapeutic strategies. It is not excluded that immunosuppression induced by chemotherapies contributed to the infection. Painful radiculoneuritis, cranial neuropathy especially facial palsy, the absence of specific antineuronal antibodies (typically anti-Ri, ANNA-2), and an important pleocytosis could suggest neuroborreliosis rather than the paraneoplastic syndrome. 7 A recent review suggested to distinguish between paraneoplastic (around 40% of cases) and "idiopathic" causes of OMAS, the latter group including parainfectious causes in which the underlying pathophysiologic mechanism could involve an immunologic activation secondary to the infectious process. 7 In our case, no improvement was observed on corticotherapy, whereas the patient recovered after antibiotic treatment. Such course was noted in most of the published cases of OMAS caused by borrelia, 2-5 thus suggesting a direct infectious mechanism.
OMAS, even in its incomplete presentation, can be a manifestation of neuroborreliosis and should be considered especially in endemic areas.

Study Funding
No targeted funding reported.