23374 real time scanning with jetson tx for skin cancer
play

23374: Real-Time Scanning with Jetson TX For Skin Cancer Detection - PowerPoint PPT Presentation

23374: Real-Time Scanning with Jetson TX For Skin Cancer Detection Michael Gielda Jeremy Massey Antmicro Iko Business Development Manager Director 1 Skin Cancer Most common form of cancer. ! Skin Cancer Rates Types of skin cancer


  1. 23374: Real-Time Scanning with Jetson TX For Skin Cancer Detection Michael Gielda Jeremy Massey Antmicro Iko Business Development Manager Director

  2. 1 Skin Cancer

  3. Most common form of cancer.

  4. ! Skin Cancer Rates

  5. Types of skin cancer include: Basal Cell Carcinoma (BCC) • Squamous Cell Carcinoma (SCC) • Malignant Melanoma (MM) •

  6. Malignant Melanoma Comparatively rare form. Highest mortality rate.

  7. Detect it early! Early surgical intervention is cheap and almost 100% effective.

  8. 2 Current Methods

  9. A Asymmetry B Border C Analysis Colour D Diameter E Evolution

  10. Dermoscopy

  11. Photography

  12. Digital Dermoscopy Hardware & Software

  13. Mapping Software

  14. Limitations of 3 Current Methods

  15. Primary: 7.2mm Secondary: 5.9mm

  16. Primary: 6.6mm Secondary: 5.1mm

  17. Accuracy Primary: 91.6% Secondary: 86.4%

  18. Accurate Affordable

  19. Depth Imaging

  20. Our patent-pending method of measuring surface features.

  21. Figure 1 Figure 2 Figure 3

  22. Figure 4 Figure 5 Figure 6

  23. Figure 7 Figure 8 Figure 9

  24. Length: 4.83mm

  25. Length: 4.83mm

  26. Length: 1.43mm

  27. Length: 1.42mm

  28. Length: 4.51mm

  29. Length: 4.53mm

  30. Accuracy > 99%

  31. Accurate Affordable

  32. Device Requirements Real-time 3D point cloud registration. • Real-time 2D de-mosaicing. • Real-time 3D mesh generation. •

  33. CUDA & Jetson TX Requirements are well suited to parallel processing and the NVIDIA Jetson TX is perfectly suited for the task.

  34. Features Data acquisition, display & storage: Depth and colour camera • connected to NVIDIA Jetson TX1/2 module. SD card for storage. • 5” LCD with capacitive touch • screen for UI. Communication: WiFi Connectivity • Wired Ethernet • Debugging via USB UART, OTG •

  35. Getting the Vision / 3D Data PixeLink camera for colour data. • (Liquid lens!) RealSense SR300 for depth data. •

  36. Challenges for Developing the Hardware Compactness • Easy Handling • Camera Integration • Heat Dissipation •

  37. Compactness Making the device portable and reducing overall weight. Dealing with spatial • arrangement between cameras and lighting. Battery placement -> serving • as a hand grip. Placement of the display. •

  38. Easy Handling Plexiglass construction makes (dis)assembling the prototype easy. Fully usable in field testing • while fastened. Particular sections of PCB • accessible for debugging. Convenient dismantling for • hardware maintenance.

  39. Camera Integration USB-FFC adapter board for more robust work with USB 3.0 connectors.

  40. Heat Dissipation Lots of heat sources – heat dissipating aluminum frame.

  41. Accurate Affordable

  42. Current methods of analysing and monitoring moles are labour intensive making them expensive.

  43. Acquisition process has been reduced from potentially hours to a few minutes.

  44. System designed with relatively low cost hardware.

  45. Accurate Affordable

  46. 4 Future

  47. Automated Analysis

  48. Sharpened Original Corrected

  49. A Asymmetry B Border C Colour D Diameter E Evolution

  50. Thank you! Reach us at: Michael Gielda: mgielda@antmicro.com Jeremy Massey: jeremy@ikocorp.com

Recommend


More recommend