Recent advances in deep learning have greatly improved the ability to generate analysis models from medical images. In particular, great attention is focused on quickly generating models of the left ventricle from cardiac magnetic resonance imaging (cMRI) to improve the diagnosis and prognosis of millions of patients. However, even state-of-the-art frameworks present challenges, such as discontinuities of the cardiac tissue and excessive jaggedness along the myocardial walls. These geometrical features are often anatomically incorrect and may lead to unrealistic results once the geometrical models are employed in computational analyses. In this research, we propose an end-to-end pipeline for a subject-specific model of the heart's left ventricle from Cine cMRI. Our novel pipeline incorporates the uncertainty originating from the segmentation methods in the estimation of cardiac indices, such as ejection fraction, myocardial volume changes, and global radial and longitudinal strain, during the cardiac cycle. First, we propose an anatomically-guided deep learning model to overcome the common segmentation challenges while preserving the advantages of state-of-the-art frameworks, such as computational efficiency, robustness, and abstraction capabilities. Our anatomically-guided neural networks include a B-spline head, which acts as a regularization layer during training. In addition, the introduction of the B-spline head contributes to achieving a robust uncertainty quantification of the left ventricle inner and outer walls. We validate our approach using human short-axis (SA) cMRI slices and later apply transfer learning to verify its generalization capabilities in swine long-axis (LA) cMRI slices. Finally, we use the SA and LA contours to build a Gaussian Process (GP) model to create inner and outer walls 3D surfaces, which are then used to compute global indices of cardiac functions. Our results show that the proposed pipeline generates anatomically consistent geometries while also providing a robust tool for quantifying uncertainty in the geometry and the derived cardiac indices.

Röjningsröseområdet är en typ av fornlämning från brons- och järnåldern, som är vanlig i de sydsvenska skogarna. Eftersom den nationella fornlämningsinventeringen avslutades i många län innan röjningsröseområdet fick fornlämningsstatus, finns det idag ett stort antal oregistrerade röjningsröseområden. Målet med denna studie är att utveckla och testa en metod som använder maskininlärning och LiDAR-data för att identifiera och georeferera röjningsröseområden, och som kan användas som ett hjälpmedel inför inventering. Metoden anpassades för att minimera antalet falska positiv och på så sätt undvika onödiga fältbesök. LiDAR-data från Hallands län omvandlades till raster innehållande terrängdata Simple Local Relief Model. En klassificeringsmodell skapades utifrån nätverksarkitekturen YOLOv8, och tränades med 1200 bilder på röjningsröseområden och 5544 bakgrundsbilder. Falska positiv minimerades genom att inte inkludera otydliga bilder i träningsdata, samt genom att använda ett gränsvärde på 70% för röjningsröseområden vid prediktion.  Modellen hade mycket goda prestandamått vid validering, med en precision på 100% och recall på 86% för röjningsröseområden. Modellen testades därefter i två områden: ett mindre område där klassificeringen kontrollerades med visuell tolkning, och ett större område som kontrollerades automatiskt genom överlapp med registrerade fornlämningar. Den genomsnittliga precisionen för de två testområdena är 52–78%, och recall var i genomsnitt 50%. Modellen har i genomsnitt en sämre förmåga att identifiera registrerade röjningsröseområden med liten area. Alla identifierade falska positiv vid prediktion innehöll grot. 298 potentiella oregistrerade röjningsröseområden upptäcktes i testområdena. Sammanfattningsvis uppvisar modellen god potential att fungera som ett verktyg för att identifiera röjningsröseområden inför vidare visuell tolkning och fältarbete. / The cairnfield is a type of archaeological site from the Bronze and Iron Ages, which is common in the forests of southern Sweden. Since the Swedish National Survey of ancient remains was completed in many Swedish counties before the cairnfields were classified as an archaeological site, there are currently many unregistered cairnfields. The goal of this study is to develop and test a method that uses machine learning and LiDAR-data to identify and georeference cairnfields, and which can be used as a tool by surveyors. The method was adapted to minimize false positives and thereby avoid unnecessary field visits. LiDAR-data from the County of Halland was converted to raster images of the DEM-derivative Simple Local Relief Model. A YOLOv8 classification model was trained with 1200 images of cairnfields and 5544 background images. False positives were minimized by excluding unclear images during annotation, and by using a threshold of 70% for clearance cairn areas during prediction.  The model’s performance metrics during validation were excellent, with 100% precision and 86% recall for the cairnfield class. The model was tested in two areas: a smaller area where prediction results were controlled with visual interpretation, and a larger area that was checked automatically against known archaeological sites. The average precision for the two test areas is 52–78%, and the average recall is 50%. The model has a lower ability to identify registered clearance cairn areas with small areas. All false positives identified during prediction contained slash. 298 potential unregistered cairnfields were discovered in the test areas. In summary, the classification model demonstrates good potential as a tool for identifying cairnfields before further visual interpretation and fieldwork.