Semantic segmentation of brain tumours is a fundamental task in medical image analysis that can help clinicians in diagnosing the patient and tracking the progression of any malignant entities. Accurate segmentation of brain lesions is essential for medical diagnosis and treatment planning. However, failure to acquire specific MRI imaging modalities can prevent applications from operating in critical situations, raising concerns about their reliability and overall trustworthiness. This paper presents a novel multi-modal approach for brain lesion segmentation that leverages information from four distinct imaging modalities while being robust to real-world scenarios of missing modalities, such as T1, T1c, T2, and FLAIR MRI of brains. Our proposed method can help address the challenges posed by artifacts in medical imagery due to data acquisition errors (such as patient motion) or a reconstruction algorithm's inability to represent the anatomy while ensuring a trade-off in accuracy. Our proposed regularization module makes it robust to these scenarios and ensures the reliability of lesion segmentation.