Recognizing Surgical Phases in Laparoscopic Cholecystectomy Procedures with Deep Learning

الموضوعات

علوم تطبيقية وتكنولوجية

الملخص

Surgical phase recognition is a task that relies on visual data to identify and classify different stages within a surgical procedure. This research proposes an approach that leverages both temporal and spatial features present in laparoscopic surgery videos to enhance recognition performance. The proposed method combines semantic segmentation features, spatial features, and temporally enriched spatial representations. The task is formulated as a multi-class classification problem, where each video frame is assigned to one of the predefined surgical phases. In recent years, interest in this field has grown significantly due to its potential to support real-time decision-making and improve surgical workflow efficiency. However, despite promising progress, a gap remains in effectively capturing the full range of semantic, spatial, and contextual information needed for ac-curate phase recognition. Prior research has commonly relied on conventional feature extractors such as ResNet50, which may be limited in modeling long-range dependencies and complex scene structures.In this research, we propose a hybrid feature extraction method that integrates multiple deep learning models to enhance the representation of surgical videos. The approach combines semantic segmentation, spatial feature extraction, and temporally rich spatial features to create a more comprehensive feature set. DeepLabV3 is used to extract se-mantic features that describe anatomical structures. ResNet50 captures spatial features,such as textures and fine-grained patterns, while the Vision Transformer (ViT) extracts temporally rich spatial features that reflect spatial relationships across the entire frame.These features are fused into a single hybrid tensor representation, combining segmentation precision, spatial detail, and temporal context.The resulting hybrid representation is passed into three temporal modeling architectures— Temporal Convolutional Network (TCN), Long Short-Term Memory (LSTM), and Attention-Based Sequential Former (ASFormer) — to predict surgical phases from frame-level video data. An ablation study was conducted using seven feature configurations to evaluate the individual and combined contributions of semantic, spatial, and transformer-based features across all temporal models.Our proposed Residual Vision and Segmentation with Temporal Convolution Network(RVS-TCN) model achieves state-of-the-art performance on the Cholec80 dataset, reach-ing an accuracy of 93.2 ± 4.5, outperforming models based on single feature extraction method. The model’s applicability in post-operative analysis suggests its potential to inform institutional policies focused on surgical training, performance evaluation, and quality monitoring.

روابط وملفات

التعريف والنوع

رقم الوثيقة

f0964638-bbba-4966-b998-8287c929baef

رقم العقد

0

نوع الوسائط

Crawler

نوع المحتوى

الرسائل العلمية

صيغة المصدر

رسائل ماجيستير

نوع الملف

pdf text

أسماء الملفات

1084056_1.pdf

بيانات النشر

ترجمة العنوان

التعرف على المراحل الجراحية في عمليات استئصال المرارة بالمنظار باستخدام التعلم العميق

ألقاب المؤلفين

[{"name_ar":"Abrar Mady","title_ar":"اعداد","title_en":"Preparation"},{"name_ar":"Yazan Abu Farha","title_ar":"اشراف","title_en":"Supervision"}]

اللغة

English

المصدر والدورية

اسم المصدر

Recognizing Surgical Phases in Laparoscopic Cholecystectomy Procedures with Deep Learning

المحتوى والصفحات

عدد الصفحات

0

كلمات الباحثين

surgical phase recognition, deep learning, hybrid feature extraction, Tempo-ral modeling, ResNet50, Vision Transformer, temporally rich spatial features, Cholec80

إشراف وإعداد

الإشراف

Yazan Abu Farha

الإعداد

Abrar Mady

الاقتباسات الببليوغرافية