An Integrated Detection and Treatment Recommendation Framework for Breast Cancer Using Convolutional Neural Networks and TOPSIS
Devdatta Basu, Sheetal Kashid, Sanjay Shamrao Pawar, Debabrata Datta · 2020
Breast cancer is the most prevalent type of cancer, accounting for 14 percent of all cancers among women in India, according to National Health Portal. Various machine learning including deep learning methods find extensive applications in the field of medicine i.e. Computer Aided Diagnosis (CAD) and one of its major examples is detection of cancerous cells. While many such systems explore the idea of classifying the region of abnormality as benign or malign, traditional treatment recommendation solely depends on the knowledge of the physician examining. This paper explores regression using Convolutional Neural Networks (CNN) to pin-point the probable abnormal region in mammograms and calculate several morphological, texture and histogram features associated with it. These features are then leveraged to recommend the next probable treatments. The recommendation is based on the extracted features and therefore constitutes a Multi-Criteria Decision Making (MCDM) problem. To implement this type of decision making rightfully, the paper makes use of the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) where the identified features are used as criteria and experts' opinion are used as alternatives to obtain ranked recommendations. The methodology proposed in the paper is capable of recommending the correct treatment with an accuracy of 81.5%. The proposed methodology would make treatment recommendation reachable even at the most remote places where advanced facilities including consultation with specialists are not easily available.