![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F656212%2F653282b3bb59a5dfb5422e9a52aacc17%2FMenzies_Method.png?generation=1593367499175442&alt=media\")
Domain knowledge is often times more important than purely ML techniques.
\n\nMenzies method has the highest Sensitivity and Accuracy compared to published methods such as ABCD rule and 7-point checklist.
\n\nMenzies Negative features (benign lesions):\n- Symmetrical pattern (colours, structure)\n- Single colour
\n\nMenzies Positive features (melanoma):\n- Blue-white veil\n- Multiple brown dots\n- Pseudopods\n- Radial streaming\n- Scar-like depigmentation\n- Multiple (5-6) colours\n- Multiple blue/grey dots\n- Broadened network
\n\n
Reference Link for a detailed list of all Algorithms used for detecting Malignant Melanomas:\n- ABCD rule\n- 7-point checklist\n- Menzies Method\n- C.A.S.H. algorithm\n- CHAOS and clues\n- BLINK algorithm\n- Triage Amalgamated Dermoscopic Algorithm (TADA)
", "messages": [ { "id": "905690", "postDate": "06/28/2020 18:07:31", "content": "Domain knowledge is often times more important than purely ML techniques.
\n\nMenzies method has the highest Sensitivity and Accuracy compared to published methods such as ABCD rule and 7-point checklist.
\n\nMenzies Negative features (benign lesions):\n- Symmetrical pattern (colours, structure)\n- Single colour
\n\nMenzies Positive features (melanoma):\n- Blue-white veil\n- Multiple brown dots\n- Pseudopods\n- Radial streaming\n- Scar-like depigmentation\n- Multiple (5-6) colours\n- Multiple blue/grey dots\n- Broadened network
\n\n
Reference Link for a detailed list of all Algorithms used for detecting Malignant Melanomas:\n- ABCD rule\n- 7-point checklist\n- Menzies Method\n- C.A.S.H. algorithm\n- CHAOS and clues\n- BLINK algorithm\n- Triage Amalgamated Dermoscopic Algorithm (TADA)
", "rawMarkdown": "**Domain knowledge** is often times more important than purely ML techniques.\n\n**Menzies method** has the [highest Sensitivity and Accuracy](https://jamanetwork.com/journals/jamadermatology/fullarticle/397889) compared to published methods such as ABCD rule and 7-point checklist.\n\nMenzies Negative features (benign lesions):\n- Symmetrical pattern (colours, structure)\n- Single colour\n\nMenzies Positive features (melanoma):\n- Blue-white veil\n- Multiple brown dots\n- Pseudopods\n- Radial streaming\n- Scar-like depigmentation\n- Multiple (5-6) colours\n- Multiple blue/grey dots\n- Broadened network\n\n\n\n\n[Reference Link](https://dermnetnz.org/cme/dermoscopy-course/other-algorithms-for-melanocytic-lesions/) for a detailed list of all Algorithms used for detecting Malignant Melanomas:\n- ABCD rule\n- 7-point checklist\n- Menzies Method\n- C.A.S.H. algorithm\n- CHAOS and clues\n- BLINK algorithm\n- Triage Amalgamated Dermoscopic Algorithm (TADA)", "votes": null }, { "id": "905845", "postDate": "06/28/2020 21:28:34", "content": "I'm confused, at the beggining you said that:\n> Domain knowledge is often times more important than purely ML techniques.
\n\nBut after that you make a list of algorithms. The article aswell is about algorithms.
\n\nWould kindly explain what did you try to share or explain?
", "rawMarkdown": "I'm confused, at the beggining you said that:\n> Domain knowledge is often times more important than purely ML techniques.\n\nBut after that you make a list of algorithms. The article aswell is about algorithms.\n\nWould kindly explain what did you try to share or explain?", "votes": null }, { "id": "906002", "postDate": "06/29/2020 02:52:32", "content": "@hiramcho Those algorithms are used by Doctors in the Demoscopy domain to diagnose MMs i.e. they are the domain knowledge (vs software / ML technique knowledge)!
\n\nSome research papers have been mentioned already in other Discussion Posts where they perform pre-processing to extract this kind of information.
\n\nFor example:\nStep1: (Optional step) Apply Color Constancy and then use Skin Segmentation to remove skin (background) and extract only the lesion\nStep2: Extract the number of colors present in the lesion, such as light-brown, dark-brown, black, yellow, pink, red, blue-white veil\nStep3: Use this new feature (number of colors), along with patient_id information, to improve the accuracy of your predictions
", "rawMarkdown": "hiramcho Those algorithms are used by Doctors in the Demoscopy **domain** to diagnose MMs i.e. they are the **domain knowledge** (vs software / ML technique knowledge)!\n\nSome research papers have been mentioned already in other Discussion Posts where they perform pre-processing to extract this kind of information.\n\nFor example:\nStep1: (Optional step) Apply [Color Constancy](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/154876) and then use **Skin Segmentation** to remove skin (background) and extract only the lesion\nStep2: Extract the **number of colors** present in the lesion, such as light-brown, dark-brown, black, yellow, pink, red, blue-white veil\nStep3: Use this **new feature** (number of colors), along with patient_id information, to improve the accuracy of your predictions", "votes": null }, { "id": "906013", "postDate": "06/29/2020 03:09:24", "content": "Then, what you tried to explain is that a hard coded solution it's possible with domain knowledge?
\n\nAs comment, i would say that domain knowledge for me (and the books \"Think Like a Data Scientist\" & \"Data Science From Scratch\") it's free of algorithms. For example me, as a molecular biologist my domain knowledge might allow me to understand the fundamental behind sequentiation but programming skills is what allow me to actually use that information.
", "rawMarkdown": "Then, what you tried to explain is that a hard coded solution it's possible with **domain knowledge**?\n\nAs comment, i would say that **domain knowledge** for me (and the books \"Think Like a Data Scientist\" & \"Data Science From Scratch\") it's free of algorithms. For example me, as a molecular biologist my domain knowledge might allow me to understand the fundamental behind sequentiation but programming skills is what allow me to actually use that information.", "votes": null }, { "id": "906509", "postDate": "06/29/2020 11:07:38", "content": "thx
", "rawMarkdown": "thx", "votes": null } ], "comments": [ { "id": 905845, "author_name": "hiramcho", "author_url": "", "post_date": "06/28/2020 21:28:34", "content": "I'm confused, at the beggining you said that:\n> Domain knowledge is often times more important than purely ML techniques.
\n\nBut after that you make a list of algorithms. The article aswell is about algorithms.
\n\nWould kindly explain what did you try to share or explain?
", "votes": null, "replies": [ { "id": 906002, "author_name": "sirishks", "author_url": "", "post_date": "06/29/2020 02:52:32", "content": "@hiramcho Those algorithms are used by Doctors in the Demoscopy domain to diagnose MMs i.e. they are the domain knowledge (vs software / ML technique knowledge)!
\n\nSome research papers have been mentioned already in other Discussion Posts where they perform pre-processing to extract this kind of information.
\n\nFor example:\nStep1: (Optional step) Apply Color Constancy and then use Skin Segmentation to remove skin (background) and extract only the lesion\nStep2: Extract the number of colors present in the lesion, such as light-brown, dark-brown, black, yellow, pink, red, blue-white veil\nStep3: Use this new feature (number of colors), along with patient_id information, to improve the accuracy of your predictions
", "votes": null, "replies": [] }, { "id": 906013, "author_name": "hiramcho", "author_url": "", "post_date": "06/29/2020 03:09:24", "content": "Then, what you tried to explain is that a hard coded solution it's possible with domain knowledge?
\n\nAs comment, i would say that domain knowledge for me (and the books \"Think Like a Data Scientist\" & \"Data Science From Scratch\") it's free of algorithms. For example me, as a molecular biologist my domain knowledge might allow me to understand the fundamental behind sequentiation but programming skills is what allow me to actually use that information.
", "votes": null, "replies": [] } ] }, { "id": 906509, "author_name": "romanweilguny", "author_url": "", "post_date": "06/29/2020 11:07:38", "content": "thx
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "905690": "**Domain knowledge** is often times more important than purely ML techniques.\n\n**Menzies method** has the [highest Sensitivity and Accuracy](https://jamanetwork.com/journals/jamadermatology/fullarticle/397889) compared to published methods such as ABCD rule and 7-point checklist.\n\nMenzies Negative features (benign lesions):\n- Symmetrical pattern (colours, structure)\n- Single colour\n\nMenzies Positive features (melanoma):\n- Blue-white veil\n- Multiple brown dots\n- Pseudopods\n- Radial streaming\n- Scar-like depigmentation\n- Multiple (5-6) colours\n- Multiple blue/grey dots\n- Broadened network\n\n\n\n\n[Reference Link](https://dermnetnz.org/cme/dermoscopy-course/other-algorithms-for-melanocytic-lesions/) for a detailed list of all Algorithms used for detecting Malignant Melanomas:\n- ABCD rule\n- 7-point checklist\n- Menzies Method\n- C.A.S.H. algorithm\n- CHAOS and clues\n- BLINK algorithm\n- Triage Amalgamated Dermoscopic Algorithm (TADA)", "905845": "I'm confused, at the beggining you said that:\n> Domain knowledge is often times more important than purely ML techniques.\n\nBut after that you make a list of algorithms. The article aswell is about algorithms.\n\nWould kindly explain what did you try to share or explain?", "906002": "hiramcho Those algorithms are used by Doctors in the Demoscopy **domain** to diagnose MMs i.e. they are the **domain knowledge** (vs software / ML technique knowledge)!\n\nSome research papers have been mentioned already in other Discussion Posts where they perform pre-processing to extract this kind of information.\n\nFor example:\nStep1: (Optional step) Apply [Color Constancy](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/154876) and then use **Skin Segmentation** to remove skin (background) and extract only the lesion\nStep2: Extract the **number of colors** present in the lesion, such as light-brown, dark-brown, black, yellow, pink, red, blue-white veil\nStep3: Use this **new feature** (number of colors), along with patient_id information, to improve the accuracy of your predictions", "906013": "Then, what you tried to explain is that a hard coded solution it's possible with **domain knowledge**?\n\nAs comment, i would say that **domain knowledge** for me (and the books \"Think Like a Data Scientist\" & \"Data Science From Scratch\") it's free of algorithms. For example me, as a molecular biologist my domain knowledge might allow me to understand the fundamental behind sequentiation but programming skills is what allow me to actually use that information.", "906509": "thx" }, "source": "meta" }