Media of University of Babylon | Master’s Thesis at the College of Information Technology Explores Improving YOLO Model Performance for small Bone Fractures Detection

Media of University of Babylon - كلية الطب

Master’s Thesis at the College of Information Technology Explores Improving YOLO Model Performance for small Bone Fractures Detection

by : Duhaa Fadill Abbas

Date : 10/7/2025 23:51:07 PM

Veiw : 13

Master’s Thesis at the College of Information Technology Explores Improving YOLO Model Performance for small Bone Fractures Detection Using Data Augmentation Based on Selective Tile

Duhaa Fadill Abbas
The Department of Software at the College of Information Technology – University of Babylon held a master s thesis defense entitled:
"Improving YOLO Model Performance for small Bone Fractures Detection Using Data Augmentation Based on Selective Tile"
presented by graduate student Muntadher Jabbar Ubaid Hassan, under the supervision of Prof. Israa Hadi Ali, on Thursday, July 10, 2025, at 9:00 a.m., in the conference hall of the college.

The thesis addressed a critical challenge in the field of medical imaging analysis: the detection of small and subtle bone fractures in X-ray images. These types of fractures are often overlooked during initial diagnoses, both by human observers and by current AI-based diagnostic systems, which may lead to delayed treatment and serious complications for patients.

The study emphasized that the rapid advancements in artificial intelligence, particularly in computer vision, present promising opportunities to develop advanced diagnostic tools that can assist healthcare professionals in making more accurate decisions. However, despite these advancements, the reliable detection of fine bone fractures remains a significant hurdle due to their small size and inconspicuous nature.

To tackle this issue, the researcher proposed a novel methodology based on Selective Small Object Tiling (SSOT). This strategy selectively identifies regions that are likely to contain subtle fractures, extracts focused image "tiles" from those areas, and enhances their visual clarity. This process trains the YOLO model to pay closer attention to fine-grained features that are typically neglected by conventional data augmentation methods.

Experimental results demonstrated the effectiveness of this approach, revealing a substantial improvement in the model’s detection performance. The research culminated in the development of a desktop application aimed at transforming this methodology into a practical tool that can be utilized by medical professionals to aid in more precise diagnosis.

This thesis exemplifies the college’s ongoing commitment to fostering applied scientific research that addresses real-world needs—particularly by harnessing artificial intelligence to enhance the quality of healthcare services.

ت	اسماء اعضاء لجنة المناقشة	اللقب العلمي	الاختصاص الدقيق	مكان العمل	المنصب
1	د. ايمان صالح صكبان	استاذ	ذكاء اصطناعي و معلوماتية حيوية	جامعة بابل / كلية تكنولوجيا المعلومات	رئيساً
2	د. ريم رزاق عبدالحسين	استاذ مساعد	ذكاء اصطناعي	جامعة تكنولوجيا المعلومات و الاتصالات / كلية معلوماتية الاعمال	عضوا
3	د. وضاح رزوقي عبود	مدرس	نظم معلومات و وسائط متعددة	جامعة بابل / كلية تكنولوجيا المعلومات	عضوا
4	د.اسراء هادي علي	استاذ	وسائط متعددة و تنقيب بيانات	جامعة بابل / كلية تكنولوجيا المعلومات	عضوا و مشرفا