Full text loading...
Practical applications of artificial intelligence in thoracic radiology
/content/chapter/10.22233/9781910443941.chap6
Practical applications of artificial intelligence in thoracic radiology
- Authors: Neil Shaw, Ben van Klinken and Tobias Schwarz
- From: BSAVA Manual of Canine and Feline Thoracic Imaging
- Item: Chapter 6, pp 97 - 101
- DOI: 10.22233/9781910443941.6
- Copyright: © 2024 British Small Animal Veterinary Association
- Publication Date: January 2024
Abstract
Artificial intelligence (AI) offers significant potential to enhance veterinary care. This chapter explores practical applications of AI in thoracic radiology. It outlines key benefits AI could provide, including reducing errors and costs as well as enabling faster assessments, as well as barriers to the adoption of AI. The unique complexities of interpreting thoracic radiographs are discussed, emphasizing the challenges involved in developing reliable AI solutions for this domain. Current capabilities of veterinary AI software are described, demonstrating functionality to automatically detect abnormalities. Finally, future trends are considered, predicting the emergence of AI-driven decision support beyond human perceptual limitations that will enable superior predictive health analysis.
/content/chapter/10.22233/9781910443941.chap6
Figures
/content/figure/10.22233/9781910443941.chap6.fig6_1
6.1
Low-dimensional representations for input data embeddings and output labels for detecting pleural effusion. The green masses represent the separation of data points that reflect normal versus abnormal anatomy. © 2024 British Small Animal Veterinary Association
10.22233/9781910443941/fig6_1_thumb.gif
10.22233/9781910443941/fig6_1.png
6.1
Low-dimensional representations for input data embeddings and output labels for detecting pleural effusion. The green masses represent the separation of data points that reflect normal versus abnormal anatomy.
/content/figure/10.22233/9781910443941.chap6.fig6_2
6.2
Translation of (a) a plain radiograph to (b) a heat map image. The pathological pleural fluid is red. © 2024 British Small Animal Veterinary Association
10.22233/9781910443941/fig6_2_thumb.gif
10.22233/9781910443941/fig6_2.png
6.2
Translation of (a) a plain radiograph to (b) a heat map image. The pathological pleural fluid is red.
/content/figure/10.22233/9781910443941.chap6.fig6_3
6.3
Thoracic radiographs analysed with the SignalPET® AI software. (a) Lateral radiograph of the neck and thorax of a cat with an obstructive oesophageal foreign body and secondary aspiration pneumonia. The AI system flags ‘Bronchial Pulmonary Pattern’, ‘Cranioventral Parenchymal Pattern’, ‘Esophageal Distension’ and ‘Esophageal Foreign Body’ as abnormal (red tabs) with a 4-out-of-4 probability index. All other listed pathological entities are listed as 4-out-of-4 normal. Interpretation by a veterinary radiologist concurs with these results. (b) Lateral thoracic radiograph of a dog with pneumothorax. The AI system flags ‘Cranioventral Parenchymal Pattern’ and ‘Pleural Gas’ as abnormal (red tabs) with a 4-out-of-4 probability index. ‘Caudodorsal Parenchymal Pattern’ is listed as normal (green) with a 2-out-of-4 probability index. All other listed pathological entities are listed as 4-out-of-4 normal. Interpretation by a veterinary radiologist concurs with the diagnosis of pneumothorax. The cranioventral and caudodorsal lung AI interpretation requires application of context by the veterinary surgeon. In pneumothorax, secondary lung collapse is a consequence of increased pleural pressure and unlikely to be an independent disease such as pneumonia. (c) Lateral thoracic radiograph of a dog with pleural effusion. The AI system flags ‘Caudodorsal Parenchymal Pattern’, ‘Cranioventral Parenchymal Pattern’, ‘Diffuse Parenchymal Pattern’ and ‘Pleural Fluid’ as abnormal (red tabs) with a 4-out-of-4 probability index and ‘Spondylosis’ as 2-out-of-4 abnormal. ‘Thoracic Mass’ is listed as normal (green) but only with a 2-out-of-4 probability index. All other listed pathological entities are listed as 4-out-of-4 normal. Interpretation by a veterinary radiologist concurs with the diagnosis of pleural effusion. Applying context, the lung changes are most likely due to effusion-related collapse and not an independent pathology. There is no obvious visible cause of the effusion, including a distinct mass. However, there is a slight dorsal deviation of the cranial thoracic trachea that would be consistent with a cranial mediastinal mass. Reflecting on the AI results is crucial to achieve a correct diagnosis or undertake further diagnostic tests if necessary. © 2024 British Small Animal Veterinary Association
10.22233/9781910443941/fig6_3_thumb.gif
10.22233/9781910443941/fig6_3.png
6.3
Thoracic radiographs analysed with the SignalPET® AI software. (a) Lateral radiograph of the neck and thorax of a cat with an obstructive oesophageal foreign body and secondary aspiration pneumonia. The AI system flags ‘Bronchial Pulmonary Pattern’, ‘Cranioventral Parenchymal Pattern’, ‘Esophageal Distension’ and ‘Esophageal Foreign Body’ as abnormal (red tabs) with a 4-out-of-4 probability index. All other listed pathological entities are listed as 4-out-of-4 normal. Interpretation by a veterinary radiologist concurs with these results. (b) Lateral thoracic radiograph of a dog with pneumothorax. The AI system flags ‘Cranioventral Parenchymal Pattern’ and ‘Pleural Gas’ as abnormal (red tabs) with a 4-out-of-4 probability index. ‘Caudodorsal Parenchymal Pattern’ is listed as normal (green) with a 2-out-of-4 probability index. All other listed pathological entities are listed as 4-out-of-4 normal. Interpretation by a veterinary radiologist concurs with the diagnosis of pneumothorax. The cranioventral and caudodorsal lung AI interpretation requires application of context by the veterinary surgeon. In pneumothorax, secondary lung collapse is a consequence of increased pleural pressure and unlikely to be an independent disease such as pneumonia. (c) Lateral thoracic radiograph of a dog with pleural effusion. The AI system flags ‘Caudodorsal Parenchymal Pattern’, ‘Cranioventral Parenchymal Pattern’, ‘Diffuse Parenchymal Pattern’ and ‘Pleural Fluid’ as abnormal (red tabs) with a 4-out-of-4 probability index and ‘Spondylosis’ as 2-out-of-4 abnormal. ‘Thoracic Mass’ is listed as normal (green) but only with a 2-out-of-4 probability index. All other listed pathological entities are listed as 4-out-of-4 normal. Interpretation by a veterinary radiologist concurs with the diagnosis of pleural effusion. Applying context, the lung changes are most likely due to effusion-related collapse and not an independent pathology. There is no obvious visible cause of the effusion, including a distinct mass. However, there is a slight dorsal deviation of the cranial thoracic trachea that would be consistent with a cranial mediastinal mass. Reflecting on the AI results is crucial to achieve a correct diagnosis or undertake further diagnostic tests if necessary.
