Reza, Md TanzimBiswas, SoumodeepNayeem, JannatulRahman, Md IftySaad, Tanim AhmedAyesha, Sabiha Salam2026-01-202026-01-2020252025-10ID 21101291ID 19301090ID 24241357ID 24241358ID 20201060http://hdl.handle.net/10361/27466Cataloged from PDF version of thesis.Includes bibliographical references (pages 58-60).This thesis is submitted in partial fulfillment of the requirements for the degree of Bachelor of Science in Computer Science, 2025.Disease diagnosis through medical image analysis using various transfer learning models and neural networks have made significant progress in recent years. However, Medical Image datasets are highly imbalanced due to the minimal number of cases of rare diseases. As a result of this imbalance, pre-trained CNN models perform poorly in detecting rare diseases in supervised classification tasks. Classes with a high number of data samples dominate in conventional supervised learning setup. Therefore, our research focused on trying to minimize the effect of this class imbalance. We proposed a hybrid architecture which put together supervised learning and few-shot learning. For common class detection, we used a pre-trained MobileNet-V2 as the base model of the classical supervised learning. For Rare classes, a few shot learning model, Relation Network was responsible for detecting rare disease classes. Our proposed hybrid architecture achieved an average of 90% F1 score on the rare classes. In contrast, we experimented with 3 pre-trained CNN models for traditional supervised learning and observed that all of them had scored poor recall or precision value with an average of 45% F1 score on the rare classes. Therefore, our findings highlighted that our proposed hybrid architectural approach is more impactful and can achieve good results. For that reason, we believe our work opens the door for researchers for future study that a hybrid approach of combining supervised and few-shot learning can be effective instead of relying on only one.69 pagesenBRAC University theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission.Supervised learningConvolutional neural networksFeature embeddingsDisease detectionMedical imagingImage analysisDiagnostic imaging--Data processing.Rare diseases--Diagnosis--Data processing.Imaging systems in medicine.Image processingRadiography, Medical.Computer vision.Neural networks (Computer science).Thesis