Tumour Detection and Segmentation in MRI Scans of the Gut Area

A tumour is a dangerous ailment formed when abnormal cells group together to form unwanted tissues. Over the past years, many have suffered greatly and some even died as a result of late detection of tumour tissues growing in them. In this paper, a means of automatically detecting and mapping out tumour-infected regions in gut MRI scans of patients is proposed with the aid of artificial neural networks and computational algorithms. The dataset used in this paper comprises 38,496 MRI 16-bit greyscale scan slices of the gut area of various patients. Each scan represents a slice of the gut area of a patient, a single patient may have multiple scans of various slices of their gut area. Here, the means of improving model robustness via data augmentation is devised, alongside a suitable metric function for the estimation of loss and accuracy as well as gradient computation in model training were discussed. The methodology proposed yields a model that achieved an average accuracy of 89–90% on inference data.